dirstate-v2: add devel config option to control write behavior

This will help us to write predictable tests checking behavior in each case.

mercurial/configitems.py

@@ -1,2893 +1,2902 b''
 # configitems.py - centralized declaration of configuration option
 #
 #  Copyright 2017 Pierre-Yves David <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 import functools
 import re
 
 from . import (
     encoding,
     error,
 )
 
 
 def loadconfigtable(ui, extname, configtable):
     """update config item known to the ui with the extension ones"""
     for section, items in sorted(configtable.items()):
         knownitems = ui._knownconfig.setdefault(section, itemregister())
         knownkeys = set(knownitems)
         newkeys = set(items)
         for key in sorted(knownkeys & newkeys):
             msg = b"extension '%s' overwrite config item '%s.%s'"
             msg %= (extname, section, key)
             ui.develwarn(msg, config=b'warn-config')
 
         knownitems.update(items)
 
 
 class configitem:
     """represent a known config item
 
     :section: the official config section where to find this item,
        :name: the official name within the section,
     :default: default value for this item,
     :alias: optional list of tuples as alternatives,
     :generic: this is a generic definition, match name using regular expression.
     """
 
     def __init__(
         self,
         section,
         name,
         default=None,
         alias=(),
         generic=False,
         priority=0,
         experimental=False,
     ):
         self.section = section
         self.name = name
         self.default = default
         self.alias = list(alias)
         self.generic = generic
         self.priority = priority
         self.experimental = experimental
         self._re = None
         if generic:
             self._re = re.compile(self.name)
 
 
 class itemregister(dict):
     """A specialized dictionary that can handle wild-card selection"""
 
     def __init__(self):
         super(itemregister, self).__init__()
         self._generics = set()
 
     def update(self, other):
         super(itemregister, self).update(other)
         self._generics.update(other._generics)
 
     def __setitem__(self, key, item):
         super(itemregister, self).__setitem__(key, item)
         if item.generic:
             self._generics.add(item)
 
     def get(self, key):
         baseitem = super(itemregister, self).get(key)
         if baseitem is not None and not baseitem.generic:
             return baseitem
 
         # search for a matching generic item
         generics = sorted(self._generics, key=(lambda x: (x.priority, x.name)))
         for item in generics:
             # we use 'match' instead of 'search' to make the matching simpler
             # for people unfamiliar with regular expression. Having the match
             # rooted to the start of the string will produce less surprising
             # result for user writing simple regex for sub-attribute.
             #
             # For example using "color\..*" match produces an unsurprising
             # result, while using search could suddenly match apparently
             # unrelated configuration that happens to contains "color."
             # anywhere. This is a tradeoff where we favor requiring ".*" on
             # some match to avoid the need to prefix most pattern with "^".
             # The "^" seems more error prone.
             if item._re.match(key):
                 return item
 
         return None
 
 
 coreitems = {}
 
 
 def _register(configtable, *args, **kwargs):
     item = configitem(*args, **kwargs)
     section = configtable.setdefault(item.section, itemregister())
     if item.name in section:
         msg = b"duplicated config item registration for '%s.%s'"
         raise error.ProgrammingError(msg % (item.section, item.name))
     section[item.name] = item
 
 
 # special value for case where the default is derived from other values
 dynamicdefault = object()
 
 # Registering actual config items
 
 
 def getitemregister(configtable):
     f = functools.partial(_register, configtable)
     # export pseudo enum as configitem.*
     f.dynamicdefault = dynamicdefault
     return f
 
 
 coreconfigitem = getitemregister(coreitems)
 
 
 def _registerdiffopts(section, configprefix=b''):
     coreconfigitem(
         section,
         configprefix + b'nodates',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'showfunc',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'unified',
         default=None,
     )
     coreconfigitem(
         section,
         configprefix + b'git',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'ignorews',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'ignorewsamount',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'ignoreblanklines',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'ignorewseol',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'nobinary',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'noprefix',
         default=False,
     )
     coreconfigitem(
         section,
         configprefix + b'word-diff',
         default=False,
     )
 
 
 coreconfigitem(
     b'alias',
     b'.*',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'auth',
     b'cookiefile',
     default=None,
 )
 _registerdiffopts(section=b'annotate')
 # bookmarks.pushing: internal hack for discovery
 coreconfigitem(
     b'bookmarks',
     b'pushing',
     default=list,
 )
 # bundle.mainreporoot: internal hack for bundlerepo
 coreconfigitem(
     b'bundle',
     b'mainreporoot',
     default=b'',
 )
 coreconfigitem(
     b'censor',
     b'policy',
     default=b'abort',
     experimental=True,
 )
 coreconfigitem(
     b'chgserver',
     b'idletimeout',
     default=3600,
 )
 coreconfigitem(
     b'chgserver',
     b'skiphash',
     default=False,
 )
 coreconfigitem(
     b'cmdserver',
     b'log',
     default=None,
 )
 coreconfigitem(
     b'cmdserver',
     b'max-log-files',
     default=7,
 )
 coreconfigitem(
     b'cmdserver',
     b'max-log-size',
     default=b'1 MB',
 )
 coreconfigitem(
     b'cmdserver',
     b'max-repo-cache',
     default=0,
     experimental=True,
 )
 coreconfigitem(
     b'cmdserver',
     b'message-encodings',
     default=list,
 )
 coreconfigitem(
     b'cmdserver',
     b'track-log',
     default=lambda: [b'chgserver', b'cmdserver', b'repocache'],
 )
 coreconfigitem(
     b'cmdserver',
     b'shutdown-on-interrupt',
     default=True,
 )
 coreconfigitem(
     b'color',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'color',
     b'mode',
     default=b'auto',
 )
 coreconfigitem(
     b'color',
     b'pagermode',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'command-templates',
     b'graphnode',
     default=None,
     alias=[(b'ui', b'graphnodetemplate')],
 )
 coreconfigitem(
     b'command-templates',
     b'log',
     default=None,
     alias=[(b'ui', b'logtemplate')],
 )
 coreconfigitem(
     b'command-templates',
     b'mergemarker',
     default=(
         b'{node|short} '
         b'{ifeq(tags, "tip", "", '
         b'ifeq(tags, "", "", "{tags} "))}'
         b'{if(bookmarks, "{bookmarks} ")}'
         b'{ifeq(branch, "default", "", "{branch} ")}'
         b'- {author|user}: {desc|firstline}'
     ),
     alias=[(b'ui', b'mergemarkertemplate')],
 )
 coreconfigitem(
     b'command-templates',
     b'pre-merge-tool-output',
     default=None,
     alias=[(b'ui', b'pre-merge-tool-output-template')],
 )
 coreconfigitem(
     b'command-templates',
     b'oneline-summary',
     default=None,
 )
 coreconfigitem(
     b'command-templates',
     b'oneline-summary.*',
     default=dynamicdefault,
     generic=True,
 )
 _registerdiffopts(section=b'commands', configprefix=b'commit.interactive.')
 coreconfigitem(
     b'commands',
     b'commit.post-status',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'grep.all-files',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'commands',
     b'merge.require-rev',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'push.require-revs',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'resolve.confirm',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'resolve.explicit-re-merge',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'resolve.mark-check',
     default=b'none',
 )
 _registerdiffopts(section=b'commands', configprefix=b'revert.interactive.')
 coreconfigitem(
     b'commands',
     b'show.aliasprefix',
     default=list,
 )
 coreconfigitem(
     b'commands',
     b'status.relative',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'status.skipstates',
     default=[],
     experimental=True,
 )
 coreconfigitem(
     b'commands',
     b'status.terse',
     default=b'',
 )
 coreconfigitem(
     b'commands',
     b'status.verbose',
     default=False,
 )
 coreconfigitem(
     b'commands',
     b'update.check',
     default=None,
 )
 coreconfigitem(
     b'commands',
     b'update.requiredest',
     default=False,
 )
 coreconfigitem(
     b'committemplate',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'convert',
     b'bzr.saverev',
     default=True,
 )
 coreconfigitem(
     b'convert',
     b'cvsps.cache',
     default=True,
 )
 coreconfigitem(
     b'convert',
     b'cvsps.fuzz',
     default=60,
 )
 coreconfigitem(
     b'convert',
     b'cvsps.logencoding',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'cvsps.mergefrom',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'cvsps.mergeto',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'git.committeractions',
     default=lambda: [b'messagedifferent'],
 )
 coreconfigitem(
     b'convert',
     b'git.extrakeys',
     default=list,
 )
 coreconfigitem(
     b'convert',
     b'git.findcopiesharder',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'git.remoteprefix',
     default=b'remote',
 )
 coreconfigitem(
     b'convert',
     b'git.renamelimit',
     default=400,
 )
 coreconfigitem(
     b'convert',
     b'git.saverev',
     default=True,
 )
 coreconfigitem(
     b'convert',
     b'git.similarity',
     default=50,
 )
 coreconfigitem(
     b'convert',
     b'git.skipsubmodules',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'hg.clonebranches',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'hg.ignoreerrors',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'hg.preserve-hash',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'hg.revs',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'hg.saverev',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'hg.sourcename',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'hg.startrev',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'hg.tagsbranch',
     default=b'default',
 )
 coreconfigitem(
     b'convert',
     b'hg.usebranchnames',
     default=True,
 )
 coreconfigitem(
     b'convert',
     b'ignoreancestorcheck',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'convert',
     b'localtimezone',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'p4.encoding',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'convert',
     b'p4.startrev',
     default=0,
 )
 coreconfigitem(
     b'convert',
     b'skiptags',
     default=False,
 )
 coreconfigitem(
     b'convert',
     b'svn.debugsvnlog',
     default=True,
 )
 coreconfigitem(
     b'convert',
     b'svn.trunk',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'svn.tags',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'svn.branches',
     default=None,
 )
 coreconfigitem(
     b'convert',
     b'svn.startrev',
     default=0,
 )
 coreconfigitem(
     b'convert',
     b'svn.dangerous-set-commit-dates',
     default=False,
 )
 coreconfigitem(
     b'debug',
     b'dirstate.delaywrite',
     default=0,
 )
 coreconfigitem(
     b'debug',
     b'revlog.verifyposition.changelog',
     default=b'',
 )
 coreconfigitem(
     b'debug',
     b'revlog.debug-delta',
     default=False,
 )
 coreconfigitem(
     b'defaults',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'devel',
     b'all-warnings',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'bundle2.debug',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'bundle.delta',
     default=b'',
 )
 coreconfigitem(
     b'devel',
     b'cache-vfs',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'check-locks',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'check-relroot',
     default=False,
 )
 # Track copy information for all file, not just "added" one (very slow)
 coreconfigitem(
     b'devel',
     b'copy-tracing.trace-all-files',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'default-date',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'deprec-warn',
     default=False,
 )
+# possible values:
+# - auto (the default)
+# - force-append
+# - force-new
+coreconfigitem(
+    b'devel',
+    b'dirstate.v2.data_update_mode',
+    default="auto",
+)
 coreconfigitem(
     b'devel',
     b'disableloaddefaultcerts',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'warn-empty-changegroup',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'legacy.exchange',
     default=list,
 )
 # When True, revlogs use a special reference version of the nodemap, that is not
 # performant but is "known" to behave properly.
 coreconfigitem(
     b'devel',
     b'persistent-nodemap',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'servercafile',
     default=b'',
 )
 coreconfigitem(
     b'devel',
     b'serverexactprotocol',
     default=b'',
 )
 coreconfigitem(
     b'devel',
     b'serverrequirecert',
     default=False,
 )
 # Makes the status algorithm wait for the existence of this file
 # (or until a timeout of `devel.sync.status.pre-dirstate-write-file-timeout`
 # seconds) before taking the lock and writing the dirstate.
 # Status signals that it's ready to wait by creating a file
 # with the same name + `.waiting`.
 # Useful when testing race conditions.
 coreconfigitem(
     b'devel',
     b'sync.status.pre-dirstate-write-file',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'sync.status.pre-dirstate-write-file-timeout',
     default=2,
 )
 coreconfigitem(
     b'devel',
     b'strip-obsmarkers',
     default=True,
 )
 coreconfigitem(
     b'devel',
     b'warn-config',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'warn-config-default',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'user.obsmarker',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'warn-config-unknown',
     default=None,
 )
 coreconfigitem(
     b'devel',
     b'debug.copies',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'copy-tracing.multi-thread',
     default=True,
 )
 coreconfigitem(
     b'devel',
     b'debug.extensions',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'debug.repo-filters',
     default=False,
 )
 coreconfigitem(
     b'devel',
     b'debug.peer-request',
     default=False,
 )
 # If discovery.exchange-heads is False, the discovery will not start with
 # remote head fetching and local head querying.
 coreconfigitem(
     b'devel',
     b'discovery.exchange-heads',
     default=True,
 )
 # If discovery.grow-sample is False, the sample size used in set discovery will
 # not be increased through the process
 coreconfigitem(
     b'devel',
     b'discovery.grow-sample',
     default=True,
 )
 # When discovery.grow-sample.dynamic is True, the default, the sample size is
 # adapted to the shape of the undecided set (it is set to the max of:
 # <target-size>, len(roots(undecided)), len(heads(undecided)
 coreconfigitem(
     b'devel',
     b'discovery.grow-sample.dynamic',
     default=True,
 )
 # discovery.grow-sample.rate control the rate at which the sample grow
 coreconfigitem(
     b'devel',
     b'discovery.grow-sample.rate',
     default=1.05,
 )
 # If discovery.randomize is False, random sampling during discovery are
 # deterministic. It is meant for integration tests.
 coreconfigitem(
     b'devel',
     b'discovery.randomize',
     default=True,
 )
 # Control the initial size of the discovery sample
 coreconfigitem(
     b'devel',
     b'discovery.sample-size',
     default=200,
 )
 # Control the initial size of the discovery for initial change
 coreconfigitem(
     b'devel',
     b'discovery.sample-size.initial',
     default=100,
 )
 _registerdiffopts(section=b'diff')
 coreconfigitem(
     b'diff',
     b'merge',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'email',
     b'bcc',
     default=None,
 )
 coreconfigitem(
     b'email',
     b'cc',
     default=None,
 )
 coreconfigitem(
     b'email',
     b'charsets',
     default=list,
 )
 coreconfigitem(
     b'email',
     b'from',
     default=None,
 )
 coreconfigitem(
     b'email',
     b'method',
     default=b'smtp',
 )
 coreconfigitem(
     b'email',
     b'reply-to',
     default=None,
 )
 coreconfigitem(
     b'email',
     b'to',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'archivemetatemplate',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'experimental',
     b'auto-publish',
     default=b'publish',
 )
 coreconfigitem(
     b'experimental',
     b'bundle-phases',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'bundle2-advertise',
     default=True,
 )
 coreconfigitem(
     b'experimental',
     b'bundle2-output-capture',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'bundle2.pushback',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'bundle2lazylocking',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecomplevel',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecomplevel.bzip2',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecomplevel.gzip',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecomplevel.none',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecomplevel.zstd',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecompthreads',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecompthreads.bzip2',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecompthreads.gzip',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecompthreads.none',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'bundlecompthreads.zstd',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'changegroup3',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'changegroup4',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'cleanup-as-archived',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'clientcompressionengines',
     default=list,
 )
 coreconfigitem(
     b'experimental',
     b'copytrace',
     default=b'on',
 )
 coreconfigitem(
     b'experimental',
     b'copytrace.movecandidateslimit',
     default=100,
 )
 coreconfigitem(
     b'experimental',
     b'copytrace.sourcecommitlimit',
     default=100,
 )
 coreconfigitem(
     b'experimental',
     b'copies.read-from',
     default=b"filelog-only",
 )
 coreconfigitem(
     b'experimental',
     b'copies.write-to',
     default=b'filelog-only',
 )
 coreconfigitem(
     b'experimental',
     b'crecordtest',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'directaccess',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'directaccess.revnums',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'editortmpinhg',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'evolution',
     default=list,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.allowdivergence',
     default=False,
     alias=[(b'experimental', b'allowdivergence')],
 )
 coreconfigitem(
     b'experimental',
     b'evolution.allowunstable',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.createmarkers',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.effect-flags',
     default=True,
     alias=[(b'experimental', b'effect-flags')],
 )
 coreconfigitem(
     b'experimental',
     b'evolution.exchange',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.bundle-obsmarker',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.bundle-obsmarker:mandatory',
     default=True,
 )
 coreconfigitem(
     b'experimental',
     b'log.topo',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.report-instabilities',
     default=True,
 )
 coreconfigitem(
     b'experimental',
     b'evolution.track-operation',
     default=True,
 )
 # repo-level config to exclude a revset visibility
 #
 # The target use case is to use `share` to expose different subset of the same
 # repository, especially server side. See also `server.view`.
 coreconfigitem(
     b'experimental',
     b'extra-filter-revs',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'maxdeltachainspan',
     default=-1,
 )
 # tracks files which were undeleted (merge might delete them but we explicitly
 # kept/undeleted them) and creates new filenodes for them
 coreconfigitem(
     b'experimental',
     b'merge-track-salvaged',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'mmapindexthreshold',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'narrow',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'nonnormalparanoidcheck',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'exportableenviron',
     default=list,
 )
 coreconfigitem(
     b'experimental',
     b'extendedheader.index',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'extendedheader.similarity',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'graphshorten',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'graphstyle.parent',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'experimental',
     b'graphstyle.missing',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'experimental',
     b'graphstyle.grandparent',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'experimental',
     b'hook-track-tags',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'httppostargs',
     default=False,
 )
 coreconfigitem(b'experimental', b'nointerrupt', default=False)
 coreconfigitem(b'experimental', b'nointerrupt-interactiveonly', default=True)
 
 coreconfigitem(
     b'experimental',
     b'obsmarkers-exchange-debug',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'remotenames',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'removeemptydirs',
     default=True,
 )
 coreconfigitem(
     b'experimental',
     b'revert.interactive.select-to-keep',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'revisions.prefixhexnode',
     default=False,
 )
 # "out of experimental" todo list.
 #
 # * include management of a persistent nodemap in the main docket
 # * enforce a "no-truncate" policy for mmap safety
 #      - for censoring operation
 #      - for stripping operation
 #      - for rollback operation
 # * proper streaming (race free) of the docket file
 # * track garbage data to evemtually allow rewriting -existing- sidedata.
 # * Exchange-wise, we will also need to do something more efficient than
 #   keeping references to the affected revlogs, especially memory-wise when
 #   rewriting sidedata.
 # * introduce a proper solution to reduce the number of filelog related files.
 # * use caching for reading sidedata (similar to what we do for data).
 # * no longer set offset=0 if sidedata_size=0 (simplify cutoff computation).
 # * Improvement to consider
 #   - avoid compression header in chunk using the default compression?
 #   - forbid "inline" compression mode entirely?
 #   - split the data offset and flag field (the 2 bytes save are mostly trouble)
 #   - keep track of uncompressed -chunk- size (to preallocate memory better)
 #   - keep track of chain base or size (probably not that useful anymore)
 coreconfigitem(
     b'experimental',
     b'revlogv2',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'revisions.disambiguatewithin',
     default=None,
 )
 coreconfigitem(
     b'experimental',
     b'rust.index',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'server.filesdata.recommended-batch-size',
     default=50000,
 )
 coreconfigitem(
     b'experimental',
     b'server.manifestdata.recommended-batch-size',
     default=100000,
 )
 coreconfigitem(
     b'experimental',
     b'server.stream-narrow-clones',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'single-head-per-branch',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'single-head-per-branch:account-closed-heads',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'single-head-per-branch:public-changes-only',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'sparse-read',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'sparse-read.density-threshold',
     default=0.50,
 )
 coreconfigitem(
     b'experimental',
     b'sparse-read.min-gap-size',
     default=b'65K',
 )
 coreconfigitem(
     b'experimental',
     b'treemanifest',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'update.atomic-file',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'web.full-garbage-collection-rate',
     default=1,  # still forcing a full collection on each request
 )
 coreconfigitem(
     b'experimental',
     b'worker.wdir-get-thread-safe',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'worker.repository-upgrade',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'xdiff',
     default=False,
 )
 coreconfigitem(
     b'extensions',
     b'[^:]*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'extensions',
     b'[^:]*:required',
     default=False,
     generic=True,
 )
 coreconfigitem(
     b'extdata',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'format',
     b'bookmarks-in-store',
     default=False,
 )
 coreconfigitem(
     b'format',
     b'chunkcachesize',
     default=None,
     experimental=True,
 )
 coreconfigitem(
     # Enable this dirstate format *when creating a new repository*.
     # Which format to use for existing repos is controlled by .hg/requires
     b'format',
     b'use-dirstate-v2',
     default=False,
     experimental=True,
     alias=[(b'format', b'exp-rc-dirstate-v2')],
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-v2.automatic-upgrade-of-mismatching-repositories',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-v2.automatic-upgrade-of-mismatching-repositories:quiet',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-tracked-hint',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-tracked-hint.version',
     default=1,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-tracked-hint.automatic-upgrade-of-mismatching-repositories',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-dirstate-tracked-hint.automatic-upgrade-of-mismatching-repositories:quiet',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'dotencode',
     default=True,
 )
 coreconfigitem(
     b'format',
     b'generaldelta',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'manifestcachesize',
     default=None,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'maxchainlen',
     default=dynamicdefault,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'obsstore-version',
     default=None,
 )
 coreconfigitem(
     b'format',
     b'sparse-revlog',
     default=True,
 )
 coreconfigitem(
     b'format',
     b'revlog-compression',
     default=lambda: [b'zstd', b'zlib'],
     alias=[(b'experimental', b'format.compression')],
 )
 # Experimental TODOs:
 #
 # * Same as for revlogv2 (but for the reduction of the number of files)
 # * Actually computing the rank of changesets
 # * Improvement to investigate
 #   - storing .hgtags fnode
 #   - storing branch related identifier
 
 coreconfigitem(
     b'format',
     b'exp-use-changelog-v2',
     default=None,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'usefncache',
     default=True,
 )
 coreconfigitem(
     b'format',
     b'usegeneraldelta',
     default=True,
 )
 coreconfigitem(
     b'format',
     b'usestore',
     default=True,
 )
 
 
 def _persistent_nodemap_default():
     """compute `use-persistent-nodemap` default value
 
     The feature is disabled unless a fast implementation is available.
     """
     from . import policy
 
     return policy.importrust('revlog') is not None
 
 
 coreconfigitem(
     b'format',
     b'use-persistent-nodemap',
     default=_persistent_nodemap_default,
 )
 coreconfigitem(
     b'format',
     b'exp-use-copies-side-data-changeset',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-share-safe',
     default=True,
 )
 coreconfigitem(
     b'format',
     b'use-share-safe.automatic-upgrade-of-mismatching-repositories',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'format',
     b'use-share-safe.automatic-upgrade-of-mismatching-repositories:quiet',
     default=False,
     experimental=True,
 )
 
 # Moving this on by default means we are confident about the scaling of phases.
 # This is not garanteed to be the case at the time this message is written.
 coreconfigitem(
     b'format',
     b'use-internal-phase',
     default=False,
     experimental=True,
 )
 # The interaction between the archived phase and obsolescence markers needs to
 # be sorted out before wider usage of this are to be considered.
 #
 # At the time this message is written, behavior when archiving obsolete
 # changeset differ significantly from stripping. As part of stripping, we also
 # remove the obsolescence marker associated to the stripped changesets,
 # revealing the precedecessors changesets when applicable. When archiving, we
 # don't touch the obsolescence markers, keeping everything hidden. This can
 # result in quite confusing situation for people combining exchanging draft
 # with the archived phases. As some markers needed by others may be skipped
 # during exchange.
 coreconfigitem(
     b'format',
     b'exp-archived-phase',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'shelve',
     b'store',
     default=b'internal',
     experimental=True,
 )
 coreconfigitem(
     b'fsmonitor',
     b'warn_when_unused',
     default=True,
 )
 coreconfigitem(
     b'fsmonitor',
     b'warn_update_file_count',
     default=50000,
 )
 coreconfigitem(
     b'fsmonitor',
     b'warn_update_file_count_rust',
     default=400000,
 )
 coreconfigitem(
     b'help',
     br'hidden-command\..*',
     default=False,
     generic=True,
 )
 coreconfigitem(
     b'help',
     br'hidden-topic\..*',
     default=False,
     generic=True,
 )
 coreconfigitem(
     b'hooks',
     b'[^:]*',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'hooks',
     b'.*:run-with-plain',
     default=True,
     generic=True,
 )
 coreconfigitem(
     b'hgweb-paths',
     b'.*',
     default=list,
     generic=True,
 )
 coreconfigitem(
     b'hostfingerprints',
     b'.*',
     default=list,
     generic=True,
 )
 coreconfigitem(
     b'hostsecurity',
     b'ciphers',
     default=None,
 )
 coreconfigitem(
     b'hostsecurity',
     b'minimumprotocol',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'hostsecurity',
     b'.*:minimumprotocol$',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'hostsecurity',
     b'.*:ciphers$',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'hostsecurity',
     b'.*:fingerprints$',
     default=list,
     generic=True,
 )
 coreconfigitem(
     b'hostsecurity',
     b'.*:verifycertsfile$',
     default=None,
     generic=True,
 )
 
 coreconfigitem(
     b'http_proxy',
     b'always',
     default=False,
 )
 coreconfigitem(
     b'http_proxy',
     b'host',
     default=None,
 )
 coreconfigitem(
     b'http_proxy',
     b'no',
     default=list,
 )
 coreconfigitem(
     b'http_proxy',
     b'passwd',
     default=None,
 )
 coreconfigitem(
     b'http_proxy',
     b'user',
     default=None,
 )
 
 coreconfigitem(
     b'http',
     b'timeout',
     default=None,
 )
 
 coreconfigitem(
     b'logtoprocess',
     b'commandexception',
     default=None,
 )
 coreconfigitem(
     b'logtoprocess',
     b'commandfinish',
     default=None,
 )
 coreconfigitem(
     b'logtoprocess',
     b'command',
     default=None,
 )
 coreconfigitem(
     b'logtoprocess',
     b'develwarn',
     default=None,
 )
 coreconfigitem(
     b'logtoprocess',
     b'uiblocked',
     default=None,
 )
 coreconfigitem(
     b'merge',
     b'checkunknown',
     default=b'abort',
 )
 coreconfigitem(
     b'merge',
     b'checkignored',
     default=b'abort',
 )
 coreconfigitem(
     b'experimental',
     b'merge.checkpathconflicts',
     default=False,
 )
 coreconfigitem(
     b'merge',
     b'followcopies',
     default=True,
 )
 coreconfigitem(
     b'merge',
     b'on-failure',
     default=b'continue',
 )
 coreconfigitem(
     b'merge',
     b'preferancestor',
     default=lambda: [b'*'],
     experimental=True,
 )
 coreconfigitem(
     b'merge',
     b'strict-capability-check',
     default=False,
 )
 coreconfigitem(
     b'merge',
     b'disable-partial-tools',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     b'.*',
     default=None,
     generic=True,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     br'.*\.patterns',
     default=dynamicdefault,
     generic=True,
     priority=-1,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     br'.*\.executable$',
     default=dynamicdefault,
     generic=True,
     priority=-1,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     br'.*\.order',
     default=0,
     generic=True,
     priority=-1,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     br'.*\.args',
     default=b"$local $base $other",
     generic=True,
     priority=-1,
     experimental=True,
 )
 coreconfigitem(
     b'partial-merge-tools',
     br'.*\.disable',
     default=False,
     generic=True,
     priority=-1,
     experimental=True,
 )
 coreconfigitem(
     b'merge-tools',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.args$',
     default=b"$local $base $other",
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.binary$',
     default=False,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.check$',
     default=list,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.checkchanged$',
     default=False,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.executable$',
     default=dynamicdefault,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.fixeol$',
     default=False,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.gui$',
     default=False,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.mergemarkers$',
     default=b'basic',
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.mergemarkertemplate$',
     default=dynamicdefault,  # take from command-templates.mergemarker
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.priority$',
     default=0,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.premerge$',
     default=dynamicdefault,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'merge-tools',
     br'.*\.symlink$',
     default=False,
     generic=True,
     priority=-1,
 )
 coreconfigitem(
     b'pager',
     b'attend-.*',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'pager',
     b'ignore',
     default=list,
 )
 coreconfigitem(
     b'pager',
     b'pager',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'patch',
     b'eol',
     default=b'strict',
 )
 coreconfigitem(
     b'patch',
     b'fuzz',
     default=2,
 )
 coreconfigitem(
     b'paths',
     b'default',
     default=None,
 )
 coreconfigitem(
     b'paths',
     b'default-push',
     default=None,
 )
 coreconfigitem(
     b'paths',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'paths',
     b'.*:bookmarks.mode',
     default='default',
     generic=True,
 )
 coreconfigitem(
     b'paths',
     b'.*:multi-urls',
     default=False,
     generic=True,
 )
 coreconfigitem(
     b'paths',
     b'.*:pushrev',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'paths',
     b'.*:pushurl',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'phases',
     b'checksubrepos',
     default=b'follow',
 )
 coreconfigitem(
     b'phases',
     b'new-commit',
     default=b'draft',
 )
 coreconfigitem(
     b'phases',
     b'publish',
     default=True,
 )
 coreconfigitem(
     b'profiling',
     b'enabled',
     default=False,
 )
 coreconfigitem(
     b'profiling',
     b'format',
     default=b'text',
 )
 coreconfigitem(
     b'profiling',
     b'freq',
     default=1000,
 )
 coreconfigitem(
     b'profiling',
     b'limit',
     default=30,
 )
 coreconfigitem(
     b'profiling',
     b'nested',
     default=0,
 )
 coreconfigitem(
     b'profiling',
     b'output',
     default=None,
 )
 coreconfigitem(
     b'profiling',
     b'showmax',
     default=0.999,
 )
 coreconfigitem(
     b'profiling',
     b'showmin',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'profiling',
     b'showtime',
     default=True,
 )
 coreconfigitem(
     b'profiling',
     b'sort',
     default=b'inlinetime',
 )
 coreconfigitem(
     b'profiling',
     b'statformat',
     default=b'hotpath',
 )
 coreconfigitem(
     b'profiling',
     b'time-track',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'profiling',
     b'type',
     default=b'stat',
 )
 coreconfigitem(
     b'progress',
     b'assume-tty',
     default=False,
 )
 coreconfigitem(
     b'progress',
     b'changedelay',
     default=1,
 )
 coreconfigitem(
     b'progress',
     b'clear-complete',
     default=True,
 )
 coreconfigitem(
     b'progress',
     b'debug',
     default=False,
 )
 coreconfigitem(
     b'progress',
     b'delay',
     default=3,
 )
 coreconfigitem(
     b'progress',
     b'disable',
     default=False,
 )
 coreconfigitem(
     b'progress',
     b'estimateinterval',
     default=60.0,
 )
 coreconfigitem(
     b'progress',
     b'format',
     default=lambda: [b'topic', b'bar', b'number', b'estimate'],
 )
 coreconfigitem(
     b'progress',
     b'refresh',
     default=0.1,
 )
 coreconfigitem(
     b'progress',
     b'width',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'pull',
     b'confirm',
     default=False,
 )
 coreconfigitem(
     b'push',
     b'pushvars.server',
     default=False,
 )
 coreconfigitem(
     b'rewrite',
     b'backup-bundle',
     default=True,
     alias=[(b'ui', b'history-editing-backup')],
 )
 coreconfigitem(
     b'rewrite',
     b'update-timestamp',
     default=False,
 )
 coreconfigitem(
     b'rewrite',
     b'empty-successor',
     default=b'skip',
     experimental=True,
 )
 # experimental as long as format.use-dirstate-v2 is.
 coreconfigitem(
     b'storage',
     b'dirstate-v2.slow-path',
     default=b"abort",
     experimental=True,
 )
 coreconfigitem(
     b'storage',
     b'new-repo-backend',
     default=b'revlogv1',
     experimental=True,
 )
 coreconfigitem(
     b'storage',
     b'revlog.optimize-delta-parent-choice',
     default=True,
     alias=[(b'format', b'aggressivemergedeltas')],
 )
 coreconfigitem(
     b'storage',
     b'revlog.issue6528.fix-incoming',
     default=True,
 )
 # experimental as long as rust is experimental (or a C version is implemented)
 coreconfigitem(
     b'storage',
     b'revlog.persistent-nodemap.mmap',
     default=True,
 )
 # experimental as long as format.use-persistent-nodemap is.
 coreconfigitem(
     b'storage',
     b'revlog.persistent-nodemap.slow-path',
     default=b"abort",
 )
 
 coreconfigitem(
     b'storage',
     b'revlog.reuse-external-delta',
     default=True,
 )
 coreconfigitem(
     b'storage',
     b'revlog.reuse-external-delta-parent',
     default=None,
 )
 coreconfigitem(
     b'storage',
     b'revlog.zlib.level',
     default=None,
 )
 coreconfigitem(
     b'storage',
     b'revlog.zstd.level',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bookmarks-pushkey-compat',
     default=True,
 )
 coreconfigitem(
     b'server',
     b'bundle1',
     default=True,
 )
 coreconfigitem(
     b'server',
     b'bundle1gd',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bundle1.pull',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bundle1gd.pull',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bundle1.push',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bundle1gd.push',
     default=None,
 )
 coreconfigitem(
     b'server',
     b'bundle2.stream',
     default=True,
     alias=[(b'experimental', b'bundle2.stream')],
 )
 coreconfigitem(
     b'server',
     b'compressionengines',
     default=list,
 )
 coreconfigitem(
     b'server',
     b'concurrent-push-mode',
     default=b'check-related',
 )
 coreconfigitem(
     b'server',
     b'disablefullbundle',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'maxhttpheaderlen',
     default=1024,
 )
 coreconfigitem(
     b'server',
     b'pullbundle',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'preferuncompressed',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'streamunbundle',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'uncompressed',
     default=True,
 )
 coreconfigitem(
     b'server',
     b'uncompressedallowsecret',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'view',
     default=b'served',
 )
 coreconfigitem(
     b'server',
     b'validate',
     default=False,
 )
 coreconfigitem(
     b'server',
     b'zliblevel',
     default=-1,
 )
 coreconfigitem(
     b'server',
     b'zstdlevel',
     default=3,
 )
 coreconfigitem(
     b'share',
     b'pool',
     default=None,
 )
 coreconfigitem(
     b'share',
     b'poolnaming',
     default=b'identity',
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-not-safe',
     default=b'abort',
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-safe',
     default=b'abort',
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-not-safe.warn',
     default=True,
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-safe.warn',
     default=True,
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-not-safe:verbose-upgrade',
     default=True,
 )
 coreconfigitem(
     b'share',
     b'safe-mismatch.source-safe:verbose-upgrade',
     default=True,
 )
 coreconfigitem(
     b'shelve',
     b'maxbackups',
     default=10,
 )
 coreconfigitem(
     b'smtp',
     b'host',
     default=None,
 )
 coreconfigitem(
     b'smtp',
     b'local_hostname',
     default=None,
 )
 coreconfigitem(
     b'smtp',
     b'password',
     default=None,
 )
 coreconfigitem(
     b'smtp',
     b'port',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'smtp',
     b'tls',
     default=b'none',
 )
 coreconfigitem(
     b'smtp',
     b'username',
     default=None,
 )
 coreconfigitem(
     b'sparse',
     b'missingwarning',
     default=True,
     experimental=True,
 )
 coreconfigitem(
     b'subrepos',
     b'allowed',
     default=dynamicdefault,  # to make backporting simpler
 )
 coreconfigitem(
     b'subrepos',
     b'hg:allowed',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'subrepos',
     b'git:allowed',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'subrepos',
     b'svn:allowed',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'templates',
     b'.*',
     default=None,
     generic=True,
 )
 coreconfigitem(
     b'templateconfig',
     b'.*',
     default=dynamicdefault,
     generic=True,
 )
 coreconfigitem(
     b'trusted',
     b'groups',
     default=list,
 )
 coreconfigitem(
     b'trusted',
     b'users',
     default=list,
 )
 coreconfigitem(
     b'ui',
     b'_usedassubrepo',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'allowemptycommit',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'archivemeta',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'askusername',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'available-memory',
     default=None,
 )
 
 coreconfigitem(
     b'ui',
     b'clonebundlefallback',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'clonebundleprefers',
     default=list,
 )
 coreconfigitem(
     b'ui',
     b'clonebundles',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'color',
     default=b'auto',
 )
 coreconfigitem(
     b'ui',
     b'commitsubrepos',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'debug',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'debugger',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'editor',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'ui',
     b'detailed-exit-code',
     default=False,
     experimental=True,
 )
 coreconfigitem(
     b'ui',
     b'fallbackencoding',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'forcecwd',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'forcemerge',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'formatdebug',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'formatjson',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'formatted',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'interactive',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'interface',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'interface.chunkselector',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'large-file-limit',
     default=10 * (2 ** 20),
 )
 coreconfigitem(
     b'ui',
     b'logblockedtimes',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'merge',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'mergemarkers',
     default=b'basic',
 )
 coreconfigitem(
     b'ui',
     b'message-output',
     default=b'stdio',
 )
 coreconfigitem(
     b'ui',
     b'nontty',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'origbackuppath',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'paginate',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'patch',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'portablefilenames',
     default=b'warn',
 )
 coreconfigitem(
     b'ui',
     b'promptecho',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'quiet',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'quietbookmarkmove',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'relative-paths',
     default=b'legacy',
 )
 coreconfigitem(
     b'ui',
     b'remotecmd',
     default=b'hg',
 )
 coreconfigitem(
     b'ui',
     b'report_untrusted',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'rollback',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'signal-safe-lock',
     default=True,
 )
 coreconfigitem(
     b'ui',
     b'slash',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'ssh',
     default=b'ssh',
 )
 coreconfigitem(
     b'ui',
     b'ssherrorhint',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'statuscopies',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'strict',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'style',
     default=b'',
 )
 coreconfigitem(
     b'ui',
     b'supportcontact',
     default=None,
 )
 coreconfigitem(
     b'ui',
     b'textwidth',
     default=78,
 )
 coreconfigitem(
     b'ui',
     b'timeout',
     default=b'600',
 )
 coreconfigitem(
     b'ui',
     b'timeout.warn',
     default=0,
 )
 coreconfigitem(
     b'ui',
     b'timestamp-output',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'traceback',
     default=False,
 )
 coreconfigitem(
     b'ui',
     b'tweakdefaults',
     default=False,
 )
 coreconfigitem(b'ui', b'username', alias=[(b'ui', b'user')])
 coreconfigitem(
     b'ui',
     b'verbose',
     default=False,
 )
 coreconfigitem(
     b'verify',
     b'skipflags',
     default=0,
 )
 coreconfigitem(
     b'web',
     b'allowbz2',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'allowgz',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'allow-pull',
     alias=[(b'web', b'allowpull')],
     default=True,
 )
 coreconfigitem(
     b'web',
     b'allow-push',
     alias=[(b'web', b'allow_push')],
     default=list,
 )
 coreconfigitem(
     b'web',
     b'allowzip',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'archivesubrepos',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'cache',
     default=True,
 )
 coreconfigitem(
     b'web',
     b'comparisoncontext',
     default=5,
 )
 coreconfigitem(
     b'web',
     b'contact',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'deny_push',
     default=list,
 )
 coreconfigitem(
     b'web',
     b'guessmime',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'hidden',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'labels',
     default=list,
 )
 coreconfigitem(
     b'web',
     b'logoimg',
     default=b'hglogo.png',
 )
 coreconfigitem(
     b'web',
     b'logourl',
     default=b'https://mercurial-scm.org/',
 )
 coreconfigitem(
     b'web',
     b'accesslog',
     default=b'-',
 )
 coreconfigitem(
     b'web',
     b'address',
     default=b'',
 )
 coreconfigitem(
     b'web',
     b'allow-archive',
     alias=[(b'web', b'allow_archive')],
     default=list,
 )
 coreconfigitem(
     b'web',
     b'allow_read',
     default=list,
 )
 coreconfigitem(
     b'web',
     b'baseurl',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'cacerts',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'certificate',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'collapse',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'csp',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'deny_read',
     default=list,
 )
 coreconfigitem(
     b'web',
     b'descend',
     default=True,
 )
 coreconfigitem(
     b'web',
     b'description',
     default=b"",
 )
 coreconfigitem(
     b'web',
     b'encoding',
     default=lambda: encoding.encoding,
 )
 coreconfigitem(
     b'web',
     b'errorlog',
     default=b'-',
 )
 coreconfigitem(
     b'web',
     b'ipv6',
     default=False,
 )
 coreconfigitem(
     b'web',
     b'maxchanges',
     default=10,
 )
 coreconfigitem(
     b'web',
     b'maxfiles',
     default=10,
 )
 coreconfigitem(
     b'web',
     b'maxshortchanges',
     default=60,
 )
 coreconfigitem(
     b'web',
     b'motd',
     default=b'',
 )
 coreconfigitem(
     b'web',
     b'name',
     default=dynamicdefault,
 )
 coreconfigitem(
     b'web',
     b'port',
     default=8000,
 )
 coreconfigitem(
     b'web',
     b'prefix',
     default=b'',
 )
 coreconfigitem(
     b'web',
     b'push_ssl',
     default=True,
 )
 coreconfigitem(
     b'web',
     b'refreshinterval',
     default=20,
 )
 coreconfigitem(
     b'web',
     b'server-header',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'static',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'staticurl',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'stripes',
     default=1,
 )
 coreconfigitem(
     b'web',
     b'style',
     default=b'paper',
 )
 coreconfigitem(
     b'web',
     b'templates',
     default=None,
 )
 coreconfigitem(
     b'web',
     b'view',
     default=b'served',
     experimental=True,
 )
 coreconfigitem(
     b'worker',
     b'backgroundclose',
     default=dynamicdefault,
 )
 # Windows defaults to a limit of 512 open files. A buffer of 128
 # should give us enough headway.
 coreconfigitem(
     b'worker',
     b'backgroundclosemaxqueue',
     default=384,
 )
 coreconfigitem(
     b'worker',
     b'backgroundcloseminfilecount',
     default=2048,
 )
 coreconfigitem(
     b'worker',
     b'backgroundclosethreadcount',
     default=4,
 )
 coreconfigitem(
     b'worker',
     b'enabled',
     default=True,
 )
 coreconfigitem(
     b'worker',
     b'numcpus',
     default=None,
 )
 
 # Rebase related configuration moved to core because other extension are doing
 # strange things. For example, shelve import the extensions to reuse some bit
 # without formally loading it.
 coreconfigitem(
     b'commands',
     b'rebase.requiredest',
     default=False,
 )
 coreconfigitem(
     b'experimental',
     b'rebaseskipobsolete',
     default=True,
 )
 coreconfigitem(
     b'rebase',
     b'singletransaction',
     default=False,
 )
 coreconfigitem(
     b'rebase',
     b'experimental.inmemory',
     default=False,
 )
 
 # This setting controls creation of a rebase_source extra field
 # during rebase. When False, no such field is created. This is
 # useful eg for incrementally converting changesets and then
 # rebasing them onto an existing repo.
 # WARNING: this is an advanced setting reserved for people who know
 # exactly what they are doing. Misuse of this setting can easily
 # result in obsmarker cycles and a vivid headache.
 coreconfigitem(
     b'rebase',
     b'store-source',
     default=True,
     experimental=True,
 )

mercurial/dirstatemap.py

@@ -1,693 +1,704 b''
 # dirstatemap.py
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 from .i18n import _
 
 from . import (
     error,
     pathutil,
     policy,
     txnutil,
     util,
 )
 
 from .dirstateutils import (
     docket as docketmod,
     v2,
 )
 
 parsers = policy.importmod('parsers')
 rustmod = policy.importrust('dirstate')
 
 propertycache = util.propertycache
 
 if rustmod is None:
     DirstateItem = parsers.DirstateItem
 else:
     DirstateItem = rustmod.DirstateItem
 
 rangemask = 0x7FFFFFFF
 
 WRITE_MODE_AUTO = 0
 WRITE_MODE_FORCE_NEW = 1
+WRITE_MODE_FORCE_APPEND = 2
 
 
 class _dirstatemapcommon:
     """
     Methods that are identical for both implementations of the dirstatemap
     class, with and without Rust extensions enabled.
     """
 
     # please pytype
 
     _map = None
     copymap = None
 
     def __init__(self, ui, opener, root, nodeconstants, use_dirstate_v2):
         self._use_dirstate_v2 = use_dirstate_v2
         self._nodeconstants = nodeconstants
         self._ui = ui
         self._opener = opener
         self._root = root
         self._filename = b'dirstate'
         self._nodelen = 20  # Also update Rust code when changing this!
         self._parents = None
         self._dirtyparents = False
         self._docket = None
+        write_mode = ui.config(b"devel", b"dirstate.v2.data_update_mode")
+        if write_mode == b"auto":
+            self._write_mode = WRITE_MODE_AUTO
+        elif write_mode == b"force-append":
+            self._write_mode = WRITE_MODE_FORCE_APPEND
+        elif write_mode == b"force-new":
+            self._write_mode = WRITE_MODE_FORCE_NEW
+        else:
+            # unknown value, fallback to default
+            self._write_mode = WRITE_MODE_AUTO
 
         # for consistent view between _pl() and _read() invocations
         self._pendingmode = None
 
     def preload(self):
         """Loads the underlying data, if it's not already loaded"""
         self._map
 
     def get(self, key, default=None):
         return self._map.get(key, default)
 
     def __len__(self):
         return len(self._map)
 
     def __iter__(self):
         return iter(self._map)
 
     def __contains__(self, key):
         return key in self._map
 
     def __getitem__(self, item):
         return self._map[item]
 
     ### disk interaction
 
     def _opendirstatefile(self):
         fp, mode = txnutil.trypending(self._root, self._opener, self._filename)
         if self._pendingmode is not None and self._pendingmode != mode:
             fp.close()
             raise error.Abort(
                 _(b'working directory state may be changed parallelly')
             )
         self._pendingmode = mode
         return fp
 
     def _readdirstatefile(self, size=-1):
         try:
             with self._opendirstatefile() as fp:
                 return fp.read(size)
         except FileNotFoundError:
             # File doesn't exist, so the current state is empty
             return b''
 
     @property
     def docket(self):
         if not self._docket:
             if not self._use_dirstate_v2:
                 raise error.ProgrammingError(
                     b'dirstate only has a docket in v2 format'
                 )
             self._docket = docketmod.DirstateDocket.parse(
                 self._readdirstatefile(), self._nodeconstants
             )
         return self._docket
 
     def write_v2_no_append(self, tr, st, meta, packed):
         old_docket = self.docket
         new_docket = docketmod.DirstateDocket.with_new_uuid(
             self.parents(), len(packed), meta
         )
         if old_docket.uuid == new_docket.uuid:
             raise error.ProgrammingError(b'dirstate docket name collision')
         data_filename = new_docket.data_filename()
         self._opener.write(data_filename, packed)
         # Write the new docket after the new data file has been
         # written. Because `st` was opened with `atomictemp=True`,
         # the actual `.hg/dirstate` file is only affected on close.
         st.write(new_docket.serialize())
         st.close()
         # Remove the old data file after the new docket pointing to
         # the new data file was written.
         if old_docket.uuid:
             data_filename = old_docket.data_filename()
             unlink = lambda _tr=None: self._opener.unlink(data_filename)
             if tr:
                 category = b"dirstate-v2-clean-" + old_docket.uuid
                 tr.addpostclose(category, unlink)
             else:
                 unlink()
         self._docket = new_docket
 
     ### reading/setting parents
 
     def parents(self):
         if not self._parents:
             if self._use_dirstate_v2:
                 self._parents = self.docket.parents
             else:
                 read_len = self._nodelen * 2
                 st = self._readdirstatefile(read_len)
                 l = len(st)
                 if l == read_len:
                     self._parents = (
                         st[: self._nodelen],
                         st[self._nodelen : 2 * self._nodelen],
                     )
                 elif l == 0:
                     self._parents = (
                         self._nodeconstants.nullid,
                         self._nodeconstants.nullid,
                     )
                 else:
                     raise error.Abort(
                         _(b'working directory state appears damaged!')
                     )
 
         return self._parents
 
 
 class dirstatemap(_dirstatemapcommon):
     """Map encapsulating the dirstate's contents.
 
     The dirstate contains the following state:
 
     - `identity` is the identity of the dirstate file, which can be used to
       detect when changes have occurred to the dirstate file.
 
     - `parents` is a pair containing the parents of the working copy. The
       parents are updated by calling `setparents`.
 
     - the state map maps filenames to tuples of (state, mode, size, mtime),
       where state is a single character representing 'normal', 'added',
       'removed', or 'merged'. It is read by treating the dirstate as a
       dict.  File state is updated by calling various methods (see each
       documentation for details):
 
       - `reset_state`,
       - `set_tracked`
       - `set_untracked`
       - `set_clean`
       - `set_possibly_dirty`
 
     - `copymap` maps destination filenames to their source filename.
 
     The dirstate also provides the following views onto the state:
 
     - `filefoldmap` is a dict mapping normalized filenames to the denormalized
       form that they appear as in the dirstate.
 
     - `dirfoldmap` is a dict mapping normalized directory names to the
       denormalized form that they appear as in the dirstate.
     """
 
     ### Core data storage and access
 
     @propertycache
     def _map(self):
         self._map = {}
         self.read()
         return self._map
 
     @propertycache
     def copymap(self):
         self.copymap = {}
         self._map
         return self.copymap
 
     def clear(self):
         self._map.clear()
         self.copymap.clear()
         self.setparents(self._nodeconstants.nullid, self._nodeconstants.nullid)
         util.clearcachedproperty(self, b"_dirs")
         util.clearcachedproperty(self, b"_alldirs")
         util.clearcachedproperty(self, b"filefoldmap")
         util.clearcachedproperty(self, b"dirfoldmap")
 
     def items(self):
         return self._map.items()
 
     # forward for python2,3 compat
     iteritems = items
 
     def debug_iter(self, all):
         """
         Return an iterator of (filename, state, mode, size, mtime) tuples
 
         `all` is unused when Rust is not enabled
         """
         for (filename, item) in self.items():
             yield (filename, item.state, item.mode, item.size, item.mtime)
 
     def keys(self):
         return self._map.keys()
 
     ### reading/setting parents
 
     def setparents(self, p1, p2, fold_p2=False):
         self._parents = (p1, p2)
         self._dirtyparents = True
         copies = {}
         if fold_p2:
             for f, s in self._map.items():
                 # Discard "merged" markers when moving away from a merge state
                 if s.p2_info:
                     source = self.copymap.pop(f, None)
                     if source:
                         copies[f] = source
                     s.drop_merge_data()
         return copies
 
     ### disk interaction
 
     def read(self):
         # ignore HG_PENDING because identity is used only for writing
         self.identity = util.filestat.frompath(
             self._opener.join(self._filename)
         )
 
         if self._use_dirstate_v2:
             if not self.docket.uuid:
                 return
             st = self._opener.read(self.docket.data_filename())
         else:
             st = self._readdirstatefile()
 
         if not st:
             return
 
         # TODO: adjust this estimate for dirstate-v2
         if util.safehasattr(parsers, b'dict_new_presized'):
             # Make an estimate of the number of files in the dirstate based on
             # its size. This trades wasting some memory for avoiding costly
             # resizes. Each entry have a prefix of 17 bytes followed by one or
             # two path names. Studies on various large-scale real-world repositories
             # found 54 bytes a reasonable upper limit for the average path names.
             # Copy entries are ignored for the sake of this estimate.
             self._map = parsers.dict_new_presized(len(st) // 71)
 
         # Python's garbage collector triggers a GC each time a certain number
         # of container objects (the number being defined by
         # gc.get_threshold()) are allocated. parse_dirstate creates a tuple
         # for each file in the dirstate. The C version then immediately marks
         # them as not to be tracked by the collector. However, this has no
         # effect on when GCs are triggered, only on what objects the GC looks
         # into. This means that O(number of files) GCs are unavoidable.
         # Depending on when in the process's lifetime the dirstate is parsed,
         # this can get very expensive. As a workaround, disable GC while
         # parsing the dirstate.
         #
         # (we cannot decorate the function directly since it is in a C module)
         if self._use_dirstate_v2:
             p = self.docket.parents
             meta = self.docket.tree_metadata
             parse_dirstate = util.nogc(v2.parse_dirstate)
             parse_dirstate(self._map, self.copymap, st, meta)
         else:
             parse_dirstate = util.nogc(parsers.parse_dirstate)
             p = parse_dirstate(self._map, self.copymap, st)
         if not self._dirtyparents:
             self.setparents(*p)
 
         # Avoid excess attribute lookups by fast pathing certain checks
         self.__contains__ = self._map.__contains__
         self.__getitem__ = self._map.__getitem__
         self.get = self._map.get
 
     def write(self, tr, st):
         if self._use_dirstate_v2:
             packed, meta = v2.pack_dirstate(self._map, self.copymap)
             self.write_v2_no_append(tr, st, meta, packed)
         else:
             packed = parsers.pack_dirstate(
                 self._map, self.copymap, self.parents()
             )
             st.write(packed)
             st.close()
         self._dirtyparents = False
 
     @propertycache
     def identity(self):
         self._map
         return self.identity
 
     ### code related to maintaining and accessing "extra" property
     # (e.g. "has_dir")
 
     def _dirs_incr(self, filename, old_entry=None):
         """increment the dirstate counter if applicable"""
         if (
             old_entry is None or old_entry.removed
         ) and "_dirs" in self.__dict__:
             self._dirs.addpath(filename)
         if old_entry is None and "_alldirs" in self.__dict__:
             self._alldirs.addpath(filename)
 
     def _dirs_decr(self, filename, old_entry=None, remove_variant=False):
         """decrement the dirstate counter if applicable"""
         if old_entry is not None:
             if "_dirs" in self.__dict__ and not old_entry.removed:
                 self._dirs.delpath(filename)
             if "_alldirs" in self.__dict__ and not remove_variant:
                 self._alldirs.delpath(filename)
         elif remove_variant and "_alldirs" in self.__dict__:
             self._alldirs.addpath(filename)
         if "filefoldmap" in self.__dict__:
             normed = util.normcase(filename)
             self.filefoldmap.pop(normed, None)
 
     @propertycache
     def filefoldmap(self):
         """Returns a dictionary mapping normalized case paths to their
         non-normalized versions.
         """
         try:
             makefilefoldmap = parsers.make_file_foldmap
         except AttributeError:
             pass
         else:
             return makefilefoldmap(
                 self._map, util.normcasespec, util.normcasefallback
             )
 
         f = {}
         normcase = util.normcase
         for name, s in self._map.items():
             if not s.removed:
                 f[normcase(name)] = name
         f[b'.'] = b'.'  # prevents useless util.fspath() invocation
         return f
 
     @propertycache
     def dirfoldmap(self):
         f = {}
         normcase = util.normcase
         for name in self._dirs:
             f[normcase(name)] = name
         return f
 
     def hastrackeddir(self, d):
         """
         Returns True if the dirstate contains a tracked (not removed) file
         in this directory.
         """
         return d in self._dirs
 
     def hasdir(self, d):
         """
         Returns True if the dirstate contains a file (tracked or removed)
         in this directory.
         """
         return d in self._alldirs
 
     @propertycache
     def _dirs(self):
         return pathutil.dirs(self._map, only_tracked=True)
 
     @propertycache
     def _alldirs(self):
         return pathutil.dirs(self._map)
 
     ### code related to manipulation of entries and copy-sources
 
     def reset_state(
         self,
         filename,
         wc_tracked=False,
         p1_tracked=False,
         p2_info=False,
         has_meaningful_mtime=True,
         parentfiledata=None,
     ):
         """Set a entry to a given state, diregarding all previous state
 
         This is to be used by the part of the dirstate API dedicated to
         adjusting the dirstate after a update/merge.
 
         note: calling this might result to no entry existing at all if the
         dirstate map does not see any point at having one for this file
         anymore.
         """
         # copy information are now outdated
         # (maybe new information should be in directly passed to this function)
         self.copymap.pop(filename, None)
 
         if not (p1_tracked or p2_info or wc_tracked):
             old_entry = self._map.get(filename)
             self._drop_entry(filename)
             self._dirs_decr(filename, old_entry=old_entry)
             return
 
         old_entry = self._map.get(filename)
         self._dirs_incr(filename, old_entry)
         entry = DirstateItem(
             wc_tracked=wc_tracked,
             p1_tracked=p1_tracked,
             p2_info=p2_info,
             has_meaningful_mtime=has_meaningful_mtime,
             parentfiledata=parentfiledata,
         )
         self._map[filename] = entry
 
     def set_tracked(self, filename):
         new = False
         entry = self.get(filename)
         if entry is None:
             self._dirs_incr(filename)
             entry = DirstateItem(
                 wc_tracked=True,
             )
 
             self._map[filename] = entry
             new = True
         elif not entry.tracked:
             self._dirs_incr(filename, entry)
             entry.set_tracked()
             self._refresh_entry(filename, entry)
             new = True
         else:
             # XXX This is probably overkill for more case, but we need this to
             # fully replace the `normallookup` call with `set_tracked` one.
             # Consider smoothing this in the future.
             entry.set_possibly_dirty()
             self._refresh_entry(filename, entry)
         return new
 
     def set_untracked(self, f):
         """Mark a file as no longer tracked in the dirstate map"""
         entry = self.get(f)
         if entry is None:
             return False
         else:
             self._dirs_decr(f, old_entry=entry, remove_variant=not entry.added)
             if not entry.p2_info:
                 self.copymap.pop(f, None)
             entry.set_untracked()
             self._refresh_entry(f, entry)
             return True
 
     def set_clean(self, filename, mode, size, mtime):
         """mark a file as back to a clean state"""
         entry = self[filename]
         size = size & rangemask
         entry.set_clean(mode, size, mtime)
         self._refresh_entry(filename, entry)
         self.copymap.pop(filename, None)
 
     def set_possibly_dirty(self, filename):
         """record that the current state of the file on disk is unknown"""
         entry = self[filename]
         entry.set_possibly_dirty()
         self._refresh_entry(filename, entry)
 
     def _refresh_entry(self, f, entry):
         """record updated state of an entry"""
         if not entry.any_tracked:
             self._map.pop(f, None)
 
     def _drop_entry(self, f):
         """remove any entry for file f
 
         This should also drop associated copy information
 
         The fact we actually need to drop it is the responsability of the caller"""
         self._map.pop(f, None)
         self.copymap.pop(f, None)
 
 
 if rustmod is not None:
 
     class dirstatemap(_dirstatemapcommon):
 
         ### Core data storage and access
 
         @propertycache
         def _map(self):
             """
             Fills the Dirstatemap when called.
             """
             # ignore HG_PENDING because identity is used only for writing
             self.identity = util.filestat.frompath(
                 self._opener.join(self._filename)
             )
 
             if self._use_dirstate_v2:
                 if self.docket.uuid:
                     # TODO: use mmap when possible
                     data = self._opener.read(self.docket.data_filename())
                 else:
                     data = b''
                 self._map = rustmod.DirstateMap.new_v2(
                     data, self.docket.data_size, self.docket.tree_metadata
                 )
                 parents = self.docket.parents
             else:
                 self._map, parents = rustmod.DirstateMap.new_v1(
                     self._readdirstatefile()
                 )
 
             if parents and not self._dirtyparents:
                 self.setparents(*parents)
 
             self.__contains__ = self._map.__contains__
             self.__getitem__ = self._map.__getitem__
             self.get = self._map.get
             return self._map
 
         @property
         def copymap(self):
             return self._map.copymap()
 
         def debug_iter(self, all):
             """
             Return an iterator of (filename, state, mode, size, mtime) tuples
 
             `all`: also include with `state == b' '` dirstate tree nodes that
             don't have an associated `DirstateItem`.
 
             """
             return self._map.debug_iter(all)
 
         def clear(self):
             self._map.clear()
             self.setparents(
                 self._nodeconstants.nullid, self._nodeconstants.nullid
             )
             util.clearcachedproperty(self, b"_dirs")
             util.clearcachedproperty(self, b"_alldirs")
             util.clearcachedproperty(self, b"dirfoldmap")
 
         def items(self):
             return self._map.items()
 
         # forward for python2,3 compat
         iteritems = items
 
         def keys(self):
             return iter(self._map)
 
         ### reading/setting parents
 
         def setparents(self, p1, p2, fold_p2=False):
             self._parents = (p1, p2)
             self._dirtyparents = True
             copies = {}
             if fold_p2:
                 copies = self._map.setparents_fixup()
             return copies
 
         ### disk interaction
 
         @propertycache
         def identity(self):
             self._map
             return self.identity
 
         def write(self, tr, st):
             if not self._use_dirstate_v2:
                 p1, p2 = self.parents()
                 packed = self._map.write_v1(p1, p2)
                 st.write(packed)
                 st.close()
                 self._dirtyparents = False
                 return
 
             # We can only append to an existing data file if there is one
-            write_mode = WRITE_MODE_AUTO
+            write_mode = self._write_mode
             if self.docket.uuid is None:
                 write_mode = WRITE_MODE_FORCE_NEW
             packed, meta, append = self._map.write_v2(write_mode)
             if append:
                 docket = self.docket
                 data_filename = docket.data_filename()
                 with self._opener(data_filename, b'r+b') as fp:
                     fp.seek(docket.data_size)
                     assert fp.tell() == docket.data_size
                     written = fp.write(packed)
                     if written is not None:  # py2 may return None
                         assert written == len(packed), (written, len(packed))
                 docket.data_size += len(packed)
                 docket.parents = self.parents()
                 docket.tree_metadata = meta
                 st.write(docket.serialize())
                 st.close()
             else:
                 self.write_v2_no_append(tr, st, meta, packed)
             # Reload from the newly-written file
             util.clearcachedproperty(self, b"_map")
             self._dirtyparents = False
 
         ### code related to maintaining and accessing "extra" property
         # (e.g. "has_dir")
 
         @propertycache
         def filefoldmap(self):
             """Returns a dictionary mapping normalized case paths to their
             non-normalized versions.
             """
             return self._map.filefoldmapasdict()
 
         def hastrackeddir(self, d):
             return self._map.hastrackeddir(d)
 
         def hasdir(self, d):
             return self._map.hasdir(d)
 
         @propertycache
         def dirfoldmap(self):
             f = {}
             normcase = util.normcase
             for name in self._map.tracked_dirs():
                 f[normcase(name)] = name
             return f
 
         ### code related to manipulation of entries and copy-sources
 
         def set_tracked(self, f):
             return self._map.set_tracked(f)
 
         def set_untracked(self, f):
             return self._map.set_untracked(f)
 
         def set_clean(self, filename, mode, size, mtime):
             self._map.set_clean(filename, mode, size, mtime)
 
         def set_possibly_dirty(self, f):
             self._map.set_possibly_dirty(f)
 
         def reset_state(
             self,
             filename,
             wc_tracked=False,
             p1_tracked=False,
             p2_info=False,
             has_meaningful_mtime=True,
             parentfiledata=None,
         ):
             return self._map.reset_state(
                 filename,
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 has_meaningful_mtime,
                 parentfiledata,
             )

rust/hg-core/src/dirstate_tree/dirstate_map.rs

@@ -1,1913 +1,1929 b''
 use bytes_cast::BytesCast;
 use micro_timer::timed;
 use std::borrow::Cow;
 use std::path::PathBuf;
 
 use super::on_disk;
 use super::on_disk::DirstateV2ParseError;
 use super::owning::OwningDirstateMap;
 use super::path_with_basename::WithBasename;
 use crate::dirstate::parsers::pack_entry;
 use crate::dirstate::parsers::packed_entry_size;
 use crate::dirstate::parsers::parse_dirstate_entries;
 use crate::dirstate::CopyMapIter;
 use crate::dirstate::DirstateV2Data;
 use crate::dirstate::ParentFileData;
 use crate::dirstate::StateMapIter;
 use crate::dirstate::TruncatedTimestamp;
 use crate::matchers::Matcher;
 use crate::utils::hg_path::{HgPath, HgPathBuf};
 use crate::DirstateEntry;
 use crate::DirstateError;
 use crate::DirstateMapError;
 use crate::DirstateParents;
 use crate::DirstateStatus;
 use crate::FastHashbrownMap as FastHashMap;
 use crate::PatternFileWarning;
 use crate::StatusError;
 use crate::StatusOptions;
 
 /// Append to an existing data file if the amount of unreachable data (not used
 /// anymore) is less than this fraction of the total amount of existing data.
 const ACCEPTABLE_UNREACHABLE_BYTES_RATIO: f32 = 0.5;
 
 #[derive(Debug, PartialEq, Eq)]
 /// Version of the on-disk format
 pub enum DirstateVersion {
     V1,
     V2,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub enum DirstateMapWriteMode {
     Auto,
     ForceNewDataFile,
+    ForceAppend,
 }
 
 #[derive(Debug)]
 pub struct DirstateMap<'on_disk> {
     /// Contents of the `.hg/dirstate` file
     pub(super) on_disk: &'on_disk [u8],
 
     pub(super) root: ChildNodes<'on_disk>,
 
     /// Number of nodes anywhere in the tree that have `.entry.is_some()`.
     pub(super) nodes_with_entry_count: u32,
 
     /// Number of nodes anywhere in the tree that have
     /// `.copy_source.is_some()`.
     pub(super) nodes_with_copy_source_count: u32,
 
     /// See on_disk::Header
     pub(super) ignore_patterns_hash: on_disk::IgnorePatternsHash,
 
     /// How many bytes of `on_disk` are not used anymore
     pub(super) unreachable_bytes: u32,
 
     /// Size of the data used to first load this `DirstateMap`. Used in case
     /// we need to write some new metadata, but no new data on disk.
     pub(super) old_data_size: usize,
 
     pub(super) dirstate_version: DirstateVersion,
+
+    /// Controlled by config option `devel.dirstate.v2.data_update_mode`
+    pub(super) write_mode: DirstateMapWriteMode,
 }
 
 /// Using a plain `HgPathBuf` of the full path from the repository root as a
 /// map key would also work: all paths in a given map have the same parent
 /// path, so comparing full paths gives the same result as comparing base
 /// names. However `HashMap` would waste time always re-hashing the same
 /// string prefix.
 pub(super) type NodeKey<'on_disk> = WithBasename<Cow<'on_disk, HgPath>>;
 
 /// Similar to `&'tree Cow<'on_disk, HgPath>`, but can also be returned
 /// for on-disk nodes that don’t actually have a `Cow` to borrow.
 #[derive(Debug)]
 pub(super) enum BorrowedPath<'tree, 'on_disk> {
     InMemory(&'tree HgPathBuf),
     OnDisk(&'on_disk HgPath),
 }
 
 #[derive(Debug)]
 pub(super) enum ChildNodes<'on_disk> {
     InMemory(FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>),
     OnDisk(&'on_disk [on_disk::Node]),
 }
 
 #[derive(Debug)]
 pub(super) enum ChildNodesRef<'tree, 'on_disk> {
     InMemory(&'tree FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>),
     OnDisk(&'on_disk [on_disk::Node]),
 }
 
 #[derive(Debug)]
 pub(super) enum NodeRef<'tree, 'on_disk> {
     InMemory(&'tree NodeKey<'on_disk>, &'tree Node<'on_disk>),
     OnDisk(&'on_disk on_disk::Node),
 }
 
 impl<'tree, 'on_disk> BorrowedPath<'tree, 'on_disk> {
     pub fn detach_from_tree(&self) -> Cow<'on_disk, HgPath> {
         match *self {
             BorrowedPath::InMemory(in_memory) => Cow::Owned(in_memory.clone()),
             BorrowedPath::OnDisk(on_disk) => Cow::Borrowed(on_disk),
         }
     }
 }
 
 impl<'tree, 'on_disk> std::ops::Deref for BorrowedPath<'tree, 'on_disk> {
     type Target = HgPath;
 
     fn deref(&self) -> &HgPath {
         match *self {
             BorrowedPath::InMemory(in_memory) => in_memory,
             BorrowedPath::OnDisk(on_disk) => on_disk,
         }
     }
 }
 
 impl Default for ChildNodes<'_> {
     fn default() -> Self {
         ChildNodes::InMemory(Default::default())
     }
 }
 
 impl<'on_disk> ChildNodes<'on_disk> {
     pub(super) fn as_ref<'tree>(
         &'tree self,
     ) -> ChildNodesRef<'tree, 'on_disk> {
         match self {
             ChildNodes::InMemory(nodes) => ChildNodesRef::InMemory(nodes),
             ChildNodes::OnDisk(nodes) => ChildNodesRef::OnDisk(nodes),
         }
     }
 
     pub(super) fn is_empty(&self) -> bool {
         match self {
             ChildNodes::InMemory(nodes) => nodes.is_empty(),
             ChildNodes::OnDisk(nodes) => nodes.is_empty(),
         }
     }
 
     fn make_mut(
         &mut self,
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
     ) -> Result<
         &mut FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>,
         DirstateV2ParseError,
     > {
         match self {
             ChildNodes::InMemory(nodes) => Ok(nodes),
             ChildNodes::OnDisk(nodes) => {
                 *unreachable_bytes +=
                     std::mem::size_of_val::<[on_disk::Node]>(nodes) as u32;
                 let nodes = nodes
                     .iter()
                     .map(|node| {
                         Ok((
                             node.path(on_disk)?,
                             node.to_in_memory_node(on_disk)?,
                         ))
                     })
                     .collect::<Result<_, _>>()?;
                 *self = ChildNodes::InMemory(nodes);
                 match self {
                     ChildNodes::InMemory(nodes) => Ok(nodes),
                     ChildNodes::OnDisk(_) => unreachable!(),
                 }
             }
         }
     }
 }
 
 impl<'tree, 'on_disk> ChildNodesRef<'tree, 'on_disk> {
     pub(super) fn get(
         &self,
         base_name: &HgPath,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<NodeRef<'tree, 'on_disk>>, DirstateV2ParseError> {
         match self {
             ChildNodesRef::InMemory(nodes) => Ok(nodes
                 .get_key_value(base_name)
                 .map(|(k, v)| NodeRef::InMemory(k, v))),
             ChildNodesRef::OnDisk(nodes) => {
                 let mut parse_result = Ok(());
                 let search_result = nodes.binary_search_by(|node| {
                     match node.base_name(on_disk) {
                         Ok(node_base_name) => node_base_name.cmp(base_name),
                         Err(e) => {
                             parse_result = Err(e);
                             // Dummy comparison result, `search_result` won’t
                             // be used since `parse_result` is an error
                             std::cmp::Ordering::Equal
                         }
                     }
                 });
                 parse_result.map(|()| {
                     search_result.ok().map(|i| NodeRef::OnDisk(&nodes[i]))
                 })
             }
         }
     }
 
     /// Iterate in undefined order
     pub(super) fn iter(
         &self,
     ) -> impl Iterator<Item = NodeRef<'tree, 'on_disk>> {
         match self {
             ChildNodesRef::InMemory(nodes) => itertools::Either::Left(
                 nodes.iter().map(|(k, v)| NodeRef::InMemory(k, v)),
             ),
             ChildNodesRef::OnDisk(nodes) => {
                 itertools::Either::Right(nodes.iter().map(NodeRef::OnDisk))
             }
         }
     }
 
     /// Iterate in parallel in undefined order
     pub(super) fn par_iter(
         &self,
     ) -> impl rayon::iter::ParallelIterator<Item = NodeRef<'tree, 'on_disk>>
     {
         use rayon::prelude::*;
         match self {
             ChildNodesRef::InMemory(nodes) => rayon::iter::Either::Left(
                 nodes.par_iter().map(|(k, v)| NodeRef::InMemory(k, v)),
             ),
             ChildNodesRef::OnDisk(nodes) => rayon::iter::Either::Right(
                 nodes.par_iter().map(NodeRef::OnDisk),
             ),
         }
     }
 
     pub(super) fn sorted(&self) -> Vec<NodeRef<'tree, 'on_disk>> {
         match self {
             ChildNodesRef::InMemory(nodes) => {
                 let mut vec: Vec<_> = nodes
                     .iter()
                     .map(|(k, v)| NodeRef::InMemory(k, v))
                     .collect();
                 fn sort_key<'a>(node: &'a NodeRef) -> &'a HgPath {
                     match node {
                         NodeRef::InMemory(path, _node) => path.base_name(),
                         NodeRef::OnDisk(_) => unreachable!(),
                     }
                 }
                 // `sort_unstable_by_key` doesn’t allow keys borrowing from the
                 // value: https://github.com/rust-lang/rust/issues/34162
                 vec.sort_unstable_by(|a, b| sort_key(a).cmp(sort_key(b)));
                 vec
             }
             ChildNodesRef::OnDisk(nodes) => {
                 // Nodes on disk are already sorted
                 nodes.iter().map(NodeRef::OnDisk).collect()
             }
         }
     }
 }
 
 impl<'tree, 'on_disk> NodeRef<'tree, 'on_disk> {
     pub(super) fn full_path(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'tree HgPath, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => Ok(path.full_path()),
             NodeRef::OnDisk(node) => node.full_path(on_disk),
         }
     }
 
     /// Returns a `BorrowedPath`, which can be turned into a `Cow<'on_disk,
     /// HgPath>` detached from `'tree`
     pub(super) fn full_path_borrowed(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<BorrowedPath<'tree, 'on_disk>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => match path.full_path() {
                 Cow::Borrowed(on_disk) => Ok(BorrowedPath::OnDisk(on_disk)),
                 Cow::Owned(in_memory) => Ok(BorrowedPath::InMemory(in_memory)),
             },
             NodeRef::OnDisk(node) => {
                 Ok(BorrowedPath::OnDisk(node.full_path(on_disk)?))
             }
         }
     }
 
     pub(super) fn base_name(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'tree HgPath, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => Ok(path.base_name()),
             NodeRef::OnDisk(node) => node.base_name(on_disk),
         }
     }
 
     pub(super) fn children(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<ChildNodesRef<'tree, 'on_disk>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => Ok(node.children.as_ref()),
             NodeRef::OnDisk(node) => {
                 Ok(ChildNodesRef::OnDisk(node.children(on_disk)?))
             }
         }
     }
 
     pub(super) fn has_copy_source(&self) -> bool {
         match self {
             NodeRef::InMemory(_path, node) => node.copy_source.is_some(),
             NodeRef::OnDisk(node) => node.has_copy_source(),
         }
     }
 
     pub(super) fn copy_source(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<&'tree HgPath>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => {
                 Ok(node.copy_source.as_ref().map(|s| &**s))
             }
             NodeRef::OnDisk(node) => node.copy_source(on_disk),
         }
     }
     /// Returns a `BorrowedPath`, which can be turned into a `Cow<'on_disk,
     /// HgPath>` detached from `'tree`
     pub(super) fn copy_source_borrowed(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<BorrowedPath<'tree, 'on_disk>>, DirstateV2ParseError>
     {
         Ok(match self {
             NodeRef::InMemory(_path, node) => {
                 node.copy_source.as_ref().map(|source| match source {
                     Cow::Borrowed(on_disk) => BorrowedPath::OnDisk(on_disk),
                     Cow::Owned(in_memory) => BorrowedPath::InMemory(in_memory),
                 })
             }
             NodeRef::OnDisk(node) => node
                 .copy_source(on_disk)?
                 .map(|source| BorrowedPath::OnDisk(source)),
         })
     }
 
     pub(super) fn entry(
         &self,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => {
                 Ok(node.data.as_entry().copied())
             }
             NodeRef::OnDisk(node) => node.entry(),
         }
     }
 
     pub(super) fn cached_directory_mtime(
         &self,
     ) -> Result<Option<TruncatedTimestamp>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => Ok(match node.data {
                 NodeData::CachedDirectory { mtime } => Some(mtime),
                 _ => None,
             }),
             NodeRef::OnDisk(node) => node.cached_directory_mtime(),
         }
     }
 
     pub(super) fn descendants_with_entry_count(&self) -> u32 {
         match self {
             NodeRef::InMemory(_path, node) => {
                 node.descendants_with_entry_count
             }
             NodeRef::OnDisk(node) => node.descendants_with_entry_count.get(),
         }
     }
 
     pub(super) fn tracked_descendants_count(&self) -> u32 {
         match self {
             NodeRef::InMemory(_path, node) => node.tracked_descendants_count,
             NodeRef::OnDisk(node) => node.tracked_descendants_count.get(),
         }
     }
 }
 
 /// Represents a file or a directory
 #[derive(Default, Debug)]
 pub(super) struct Node<'on_disk> {
     pub(super) data: NodeData,
 
     pub(super) copy_source: Option<Cow<'on_disk, HgPath>>,
 
     pub(super) children: ChildNodes<'on_disk>,
 
     /// How many (non-inclusive) descendants of this node have an entry.
     pub(super) descendants_with_entry_count: u32,
 
     /// How many (non-inclusive) descendants of this node have an entry whose
     /// state is "tracked".
     pub(super) tracked_descendants_count: u32,
 }
 
 #[derive(Debug)]
 pub(super) enum NodeData {
     Entry(DirstateEntry),
     CachedDirectory { mtime: TruncatedTimestamp },
     None,
 }
 
 impl Default for NodeData {
     fn default() -> Self {
         NodeData::None
     }
 }
 
 impl NodeData {
     fn has_entry(&self) -> bool {
         match self {
             NodeData::Entry(_) => true,
             _ => false,
         }
     }
 
     fn as_entry(&self) -> Option<&DirstateEntry> {
         match self {
             NodeData::Entry(entry) => Some(entry),
             _ => None,
         }
     }
 
     fn as_entry_mut(&mut self) -> Option<&mut DirstateEntry> {
         match self {
             NodeData::Entry(entry) => Some(entry),
             _ => None,
         }
     }
 }
 
 impl<'on_disk> DirstateMap<'on_disk> {
     pub(super) fn empty(on_disk: &'on_disk [u8]) -> Self {
         Self {
             on_disk,
             root: ChildNodes::default(),
             nodes_with_entry_count: 0,
             nodes_with_copy_source_count: 0,
             ignore_patterns_hash: [0; on_disk::IGNORE_PATTERNS_HASH_LEN],
             unreachable_bytes: 0,
             old_data_size: 0,
             dirstate_version: DirstateVersion::V1,
+            write_mode: DirstateMapWriteMode::Auto,
         }
     }
 
     #[timed]
     pub fn new_v2(
         on_disk: &'on_disk [u8],
         data_size: usize,
         metadata: &[u8],
     ) -> Result<Self, DirstateError> {
         if let Some(data) = on_disk.get(..data_size) {
             Ok(on_disk::read(data, metadata)?)
         } else {
             Err(DirstateV2ParseError::new("not enough bytes on disk").into())
         }
     }
 
     #[timed]
     pub fn new_v1(
         on_disk: &'on_disk [u8],
     ) -> Result<(Self, Option<DirstateParents>), DirstateError> {
         let mut map = Self::empty(on_disk);
         if map.on_disk.is_empty() {
             return Ok((map, None));
         }
 
         let parents = parse_dirstate_entries(
             map.on_disk,
             |path, entry, copy_source| {
                 let tracked = entry.tracked();
                 let node = Self::get_or_insert_node_inner(
                     map.on_disk,
                     &mut map.unreachable_bytes,
                     &mut map.root,
                     path,
                     WithBasename::to_cow_borrowed,
                     |ancestor| {
                         if tracked {
                             ancestor.tracked_descendants_count += 1
                         }
                         ancestor.descendants_with_entry_count += 1
                     },
                 )?;
                 assert!(
                     !node.data.has_entry(),
                     "duplicate dirstate entry in read"
                 );
                 assert!(
                     node.copy_source.is_none(),
                     "duplicate dirstate entry in read"
                 );
                 node.data = NodeData::Entry(*entry);
                 node.copy_source = copy_source.map(Cow::Borrowed);
                 map.nodes_with_entry_count += 1;
                 if copy_source.is_some() {
                     map.nodes_with_copy_source_count += 1
                 }
                 Ok(())
             },
         )?;
         let parents = Some(parents.clone());
 
         Ok((map, parents))
     }
 
     /// Assuming dirstate-v2 format, returns whether the next write should
     /// append to the existing data file that contains `self.on_disk` (true),
     /// or create a new data file from scratch (false).
     pub(super) fn write_should_append(&self) -> bool {
-        let ratio = self.unreachable_bytes as f32 / self.on_disk.len() as f32;
-        ratio < ACCEPTABLE_UNREACHABLE_BYTES_RATIO
+        match self.write_mode {
+            DirstateMapWriteMode::ForceAppend => true,
+            DirstateMapWriteMode::ForceNewDataFile => false,
+            DirstateMapWriteMode::Auto => {
+                let ratio =
+                    self.unreachable_bytes as f32 / self.on_disk.len() as f32;
+                ratio < ACCEPTABLE_UNREACHABLE_BYTES_RATIO
+            }
+        }
     }
 
     fn get_node<'tree>(
         &'tree self,
         path: &HgPath,
     ) -> Result<Option<NodeRef<'tree, 'on_disk>>, DirstateV2ParseError> {
         let mut children = self.root.as_ref();
         let mut components = path.components();
         let mut component =
             components.next().expect("expected at least one components");
         loop {
             if let Some(child) = children.get(component, self.on_disk)? {
                 if let Some(next_component) = components.next() {
                     component = next_component;
                     children = child.children(self.on_disk)?;
                 } else {
                     return Ok(Some(child));
                 }
             } else {
                 return Ok(None);
             }
         }
     }
 
     /// Returns a mutable reference to the node at `path` if it exists
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_node_mut<'tree>(
         &'tree mut self,
         path: &HgPath,
         each_ancestor: impl FnMut(&mut Node),
     ) -> Result<Option<&'tree mut Node<'on_disk>>, DirstateV2ParseError> {
         Self::get_node_mut_inner(
             self.on_disk,
             &mut self.unreachable_bytes,
             &mut self.root,
             path,
             each_ancestor,
         )
     }
 
     /// Lower-level version of `get_node_mut`.
     ///
     /// This takes `root` instead of `&mut self` so that callers can mutate
     /// other fields while the returned borrow is still valid.
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_node_mut_inner<'tree>(
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
         root: &'tree mut ChildNodes<'on_disk>,
         path: &HgPath,
         mut each_ancestor: impl FnMut(&mut Node),
     ) -> Result<Option<&'tree mut Node<'on_disk>>, DirstateV2ParseError> {
         let mut children = root;
         let mut components = path.components();
         let mut component =
             components.next().expect("expected at least one components");
         loop {
             if let Some(child) = children
                 .make_mut(on_disk, unreachable_bytes)?
                 .get_mut(component)
             {
                 if let Some(next_component) = components.next() {
                     each_ancestor(child);
                     component = next_component;
                     children = &mut child.children;
                 } else {
                     return Ok(Some(child));
                 }
             } else {
                 return Ok(None);
             }
         }
     }
 
     /// Get a mutable reference to the node at `path`, creating it if it does
     /// not exist.
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_or_insert_node<'tree, 'path>(
         &'tree mut self,
         path: &'path HgPath,
         each_ancestor: impl FnMut(&mut Node),
     ) -> Result<&'tree mut Node<'on_disk>, DirstateV2ParseError> {
         Self::get_or_insert_node_inner(
             self.on_disk,
             &mut self.unreachable_bytes,
             &mut self.root,
             path,
             WithBasename::to_cow_owned,
             each_ancestor,
         )
     }
 
     /// Lower-level version of `get_or_insert_node_inner`, which is used when
     /// parsing disk data to remove allocations for new nodes.
     fn get_or_insert_node_inner<'tree, 'path>(
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
         root: &'tree mut ChildNodes<'on_disk>,
         path: &'path HgPath,
         to_cow: impl Fn(
             WithBasename<&'path HgPath>,
         ) -> WithBasename<Cow<'on_disk, HgPath>>,
         mut each_ancestor: impl FnMut(&mut Node),
     ) -> Result<&'tree mut Node<'on_disk>, DirstateV2ParseError> {
         let mut child_nodes = root;
         let mut inclusive_ancestor_paths =
             WithBasename::inclusive_ancestors_of(path);
         let mut ancestor_path = inclusive_ancestor_paths
             .next()
             .expect("expected at least one inclusive ancestor");
         loop {
             let (_, child_node) = child_nodes
                 .make_mut(on_disk, unreachable_bytes)?
                 .raw_entry_mut()
                 .from_key(ancestor_path.base_name())
                 .or_insert_with(|| (to_cow(ancestor_path), Node::default()));
             if let Some(next) = inclusive_ancestor_paths.next() {
                 each_ancestor(child_node);
                 ancestor_path = next;
                 child_nodes = &mut child_node.children;
             } else {
                 return Ok(child_node);
             }
         }
     }
 
     fn reset_state(
         &mut self,
         filename: &HgPath,
         old_entry_opt: Option<DirstateEntry>,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parent_file_data_opt: Option<ParentFileData>,
     ) -> Result<(), DirstateError> {
         let (had_entry, was_tracked) = match old_entry_opt {
             Some(old_entry) => (true, old_entry.tracked()),
             None => (false, false),
         };
         let node = self.get_or_insert_node(filename, |ancestor| {
             if !had_entry {
                 ancestor.descendants_with_entry_count += 1;
             }
             if was_tracked {
                 if !wc_tracked {
                     ancestor.tracked_descendants_count = ancestor
                         .tracked_descendants_count
                         .checked_sub(1)
                         .expect("tracked count to be >= 0");
                 }
             } else {
                 if wc_tracked {
                     ancestor.tracked_descendants_count += 1;
                 }
             }
         })?;
 
         let v2_data = if let Some(parent_file_data) = parent_file_data_opt {
             DirstateV2Data {
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 mode_size: parent_file_data.mode_size,
                 mtime: if has_meaningful_mtime {
                     parent_file_data.mtime
                 } else {
                     None
                 },
                 ..Default::default()
             }
         } else {
             DirstateV2Data {
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 ..Default::default()
             }
         };
         node.data = NodeData::Entry(DirstateEntry::from_v2_data(v2_data));
         if !had_entry {
             self.nodes_with_entry_count += 1;
         }
         Ok(())
     }
 
     fn set_tracked(
         &mut self,
         filename: &HgPath,
         old_entry_opt: Option<DirstateEntry>,
     ) -> Result<bool, DirstateV2ParseError> {
         let was_tracked = old_entry_opt.map_or(false, |e| e.tracked());
         let had_entry = old_entry_opt.is_some();
         let tracked_count_increment = if was_tracked { 0 } else { 1 };
         let mut new = false;
 
         let node = self.get_or_insert_node(filename, |ancestor| {
             if !had_entry {
                 ancestor.descendants_with_entry_count += 1;
             }
 
             ancestor.tracked_descendants_count += tracked_count_increment;
         })?;
         if let Some(old_entry) = old_entry_opt {
             let mut e = old_entry.clone();
             if e.tracked() {
                 // XXX
                 // This is probably overkill for more case, but we need this to
                 // fully replace the `normallookup` call with `set_tracked`
                 // one. Consider smoothing this in the future.
                 e.set_possibly_dirty();
             } else {
                 new = true;
                 e.set_tracked();
             }
             node.data = NodeData::Entry(e)
         } else {
             node.data = NodeData::Entry(DirstateEntry::new_tracked());
             self.nodes_with_entry_count += 1;
             new = true;
         };
         Ok(new)
     }
 
     /// Set a node as untracked in the dirstate.
     ///
     /// It is the responsibility of the caller to remove the copy source and/or
     /// the entry itself if appropriate.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_untracked(
         &mut self,
         filename: &HgPath,
         old_entry: DirstateEntry,
     ) -> Result<(), DirstateV2ParseError> {
         let node = self
             .get_node_mut(filename, |ancestor| {
                 ancestor.tracked_descendants_count = ancestor
                     .tracked_descendants_count
                     .checked_sub(1)
                     .expect("tracked_descendants_count should be >= 0");
             })?
             .expect("node should exist");
         let mut new_entry = old_entry.clone();
         new_entry.set_untracked();
         node.data = NodeData::Entry(new_entry);
         Ok(())
     }
 
     /// Set a node as clean in the dirstate.
     ///
     /// It is the responsibility of the caller to remove the copy source.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_clean(
         &mut self,
         filename: &HgPath,
         old_entry: DirstateEntry,
         mode: u32,
         size: u32,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateError> {
         let node = self
             .get_node_mut(filename, |ancestor| {
                 if !old_entry.tracked() {
                     ancestor.tracked_descendants_count += 1;
                 }
             })?
             .expect("node should exist");
         let mut new_entry = old_entry.clone();
         new_entry.set_clean(mode, size, mtime);
         node.data = NodeData::Entry(new_entry);
         Ok(())
     }
 
     /// Set a node as possibly dirty in the dirstate.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_possibly_dirty(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         let node = self
             .get_node_mut(filename, |_ancestor| {})?
             .expect("node should exist");
         let entry = node.data.as_entry_mut().expect("entry should exist");
         entry.set_possibly_dirty();
         node.data = NodeData::Entry(*entry);
         Ok(())
     }
 
     /// Clears the cached mtime for the (potential) folder at `path`.
     pub(super) fn clear_cached_mtime(
         &mut self,
         path: &HgPath,
     ) -> Result<(), DirstateV2ParseError> {
         let node = match self.get_node_mut(path, |_ancestor| {})? {
             Some(node) => node,
             None => return Ok(()),
         };
         if let NodeData::CachedDirectory { .. } = &node.data {
             node.data = NodeData::None
         }
         Ok(())
     }
 
     /// Sets the cached mtime for the (potential) folder at `path`.
     pub(super) fn set_cached_mtime(
         &mut self,
         path: &HgPath,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateV2ParseError> {
         let node = match self.get_node_mut(path, |_ancestor| {})? {
             Some(node) => node,
             None => return Ok(()),
         };
         match &node.data {
             NodeData::Entry(_) => {} // Don’t overwrite an entry
             NodeData::CachedDirectory { .. } | NodeData::None => {
                 node.data = NodeData::CachedDirectory { mtime }
             }
         }
         Ok(())
     }
 
     fn iter_nodes<'tree>(
         &'tree self,
     ) -> impl Iterator<
         Item = Result<NodeRef<'tree, 'on_disk>, DirstateV2ParseError>,
     > + 'tree {
         // Depth first tree traversal.
         //
         // If we could afford internal iteration and recursion,
         // this would look like:
         //
         // ```
         // fn traverse_children(
         //     children: &ChildNodes,
         //     each: &mut impl FnMut(&Node),
         // ) {
         //     for child in children.values() {
         //         traverse_children(&child.children, each);
         //         each(child);
         //     }
         // }
         // ```
         //
         // However we want an external iterator and therefore can’t use the
         // call stack. Use an explicit stack instead:
         let mut stack = Vec::new();
         let mut iter = self.root.as_ref().iter();
         std::iter::from_fn(move || {
             while let Some(child_node) = iter.next() {
                 let children = match child_node.children(self.on_disk) {
                     Ok(children) => children,
                     Err(error) => return Some(Err(error)),
                 };
                 // Pseudo-recursion
                 let new_iter = children.iter();
                 let old_iter = std::mem::replace(&mut iter, new_iter);
                 stack.push((child_node, old_iter));
             }
             // Found the end of a `children.iter()` iterator.
             if let Some((child_node, next_iter)) = stack.pop() {
                 // "Return" from pseudo-recursion by restoring state from the
                 // explicit stack
                 iter = next_iter;
 
                 Some(Ok(child_node))
             } else {
                 // Reached the bottom of the stack, we’re done
                 None
             }
         })
     }
 
     fn count_dropped_path(unreachable_bytes: &mut u32, path: &Cow<HgPath>) {
         if let Cow::Borrowed(path) = path {
             *unreachable_bytes += path.len() as u32
         }
     }
+
+    pub(crate) fn set_write_mode(&mut self, write_mode: DirstateMapWriteMode) {
+        self.write_mode = write_mode;
+    }
 }
 
 /// Like `Iterator::filter_map`, but over a fallible iterator of `Result`s.
 ///
 /// The callback is only called for incoming `Ok` values. Errors are passed
 /// through as-is. In order to let it use the `?` operator the callback is
 /// expected to return a `Result` of `Option`, instead of an `Option` of
 /// `Result`.
 fn filter_map_results<'a, I, F, A, B, E>(
     iter: I,
     f: F,
 ) -> impl Iterator<Item = Result<B, E>> + 'a
 where
     I: Iterator<Item = Result<A, E>> + 'a,
     F: Fn(A) -> Result<Option<B>, E> + 'a,
 {
     iter.filter_map(move |result| match result {
         Ok(node) => f(node).transpose(),
         Err(e) => Some(Err(e)),
     })
 }
 
 impl OwningDirstateMap {
     pub fn clear(&mut self) {
         self.with_dmap_mut(|map| {
             map.root = Default::default();
             map.nodes_with_entry_count = 0;
             map.nodes_with_copy_source_count = 0;
         });
     }
 
     pub fn set_tracked(
         &mut self,
         filename: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         let old_entry_opt = self.get(filename)?;
         self.with_dmap_mut(|map| map.set_tracked(filename, old_entry_opt))
     }
 
     pub fn set_untracked(
         &mut self,
         filename: &HgPath,
     ) -> Result<bool, DirstateError> {
         let old_entry_opt = self.get(filename)?;
         match old_entry_opt {
             None => Ok(false),
             Some(old_entry) => {
                 if !old_entry.tracked() {
                     // `DirstateMap::set_untracked` is not a noop if
                     // already not tracked as it will decrement the
                     // tracked counters while going down.
                     return Ok(true);
                 }
                 if old_entry.added() {
                     // Untracking an "added" entry will just result in a
                     // worthless entry (and other parts of the code will
                     // complain about it), just drop it entirely.
                     self.drop_entry_and_copy_source(filename)?;
                     return Ok(true);
                 }
                 if !old_entry.p2_info() {
                     self.copy_map_remove(filename)?;
                 }
 
                 self.with_dmap_mut(|map| {
                     map.set_untracked(filename, old_entry)?;
                     Ok(true)
                 })
             }
         }
     }
 
     pub fn set_clean(
         &mut self,
         filename: &HgPath,
         mode: u32,
         size: u32,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateError> {
         let old_entry = match self.get(filename)? {
             None => {
                 return Err(
                     DirstateMapError::PathNotFound(filename.into()).into()
                 )
             }
             Some(e) => e,
         };
         self.copy_map_remove(filename)?;
         self.with_dmap_mut(|map| {
             map.set_clean(filename, old_entry, mode, size, mtime)
         })
     }
 
     pub fn set_possibly_dirty(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         if self.get(filename)?.is_none() {
             return Err(DirstateMapError::PathNotFound(filename.into()).into());
         }
         self.with_dmap_mut(|map| map.set_possibly_dirty(filename))
     }
 
     pub fn reset_state(
         &mut self,
         filename: &HgPath,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parent_file_data_opt: Option<ParentFileData>,
     ) -> Result<(), DirstateError> {
         if !(p1_tracked || p2_info || wc_tracked) {
             self.drop_entry_and_copy_source(filename)?;
             return Ok(());
         }
         self.copy_map_remove(filename)?;
         let old_entry_opt = self.get(filename)?;
         self.with_dmap_mut(|map| {
             map.reset_state(
                 filename,
                 old_entry_opt,
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 has_meaningful_mtime,
                 parent_file_data_opt,
             )
         })
     }
 
     pub fn drop_entry_and_copy_source(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         let was_tracked = self.get(filename)?.map_or(false, |e| e.tracked());
         struct Dropped {
             was_tracked: bool,
             had_entry: bool,
             had_copy_source: bool,
         }
 
         /// If this returns `Ok(Some((dropped, removed)))`, then
         ///
         /// * `dropped` is about the leaf node that was at `filename`
         /// * `removed` is whether this particular level of recursion just
         ///   removed a node in `nodes`.
         fn recur<'on_disk>(
             on_disk: &'on_disk [u8],
             unreachable_bytes: &mut u32,
             nodes: &mut ChildNodes<'on_disk>,
             path: &HgPath,
         ) -> Result<Option<(Dropped, bool)>, DirstateV2ParseError> {
             let (first_path_component, rest_of_path) =
                 path.split_first_component();
             let nodes = nodes.make_mut(on_disk, unreachable_bytes)?;
             let node = if let Some(node) = nodes.get_mut(first_path_component)
             {
                 node
             } else {
                 return Ok(None);
             };
             let dropped;
             if let Some(rest) = rest_of_path {
                 if let Some((d, removed)) = recur(
                     on_disk,
                     unreachable_bytes,
                     &mut node.children,
                     rest,
                 )? {
                     dropped = d;
                     if dropped.had_entry {
                         node.descendants_with_entry_count = node
                             .descendants_with_entry_count
                             .checked_sub(1)
                             .expect(
                                 "descendants_with_entry_count should be >= 0",
                             );
                     }
                     if dropped.was_tracked {
                         node.tracked_descendants_count = node
                             .tracked_descendants_count
                             .checked_sub(1)
                             .expect(
                                 "tracked_descendants_count should be >= 0",
                             );
                     }
 
                     // Directory caches must be invalidated when removing a
                     // child node
                     if removed {
                         if let NodeData::CachedDirectory { .. } = &node.data {
                             node.data = NodeData::None
                         }
                     }
                 } else {
                     return Ok(None);
                 }
             } else {
                 let entry = node.data.as_entry();
                 let was_tracked = entry.map_or(false, |entry| entry.tracked());
                 let had_entry = entry.is_some();
                 if had_entry {
                     node.data = NodeData::None
                 }
                 let mut had_copy_source = false;
                 if let Some(source) = &node.copy_source {
                     DirstateMap::count_dropped_path(unreachable_bytes, source);
                     had_copy_source = true;
                     node.copy_source = None
                 }
                 dropped = Dropped {
                     was_tracked,
                     had_entry,
                     had_copy_source,
                 };
             }
             // After recursion, for both leaf (rest_of_path is None) nodes and
             // parent nodes, remove a node if it just became empty.
             let remove = !node.data.has_entry()
                 && node.copy_source.is_none()
                 && node.children.is_empty();
             if remove {
                 let (key, _) =
                     nodes.remove_entry(first_path_component).unwrap();
                 DirstateMap::count_dropped_path(
                     unreachable_bytes,
                     key.full_path(),
                 )
             }
             Ok(Some((dropped, remove)))
         }
 
         self.with_dmap_mut(|map| {
             if let Some((dropped, _removed)) = recur(
                 map.on_disk,
                 &mut map.unreachable_bytes,
                 &mut map.root,
                 filename,
             )? {
                 if dropped.had_entry {
                     map.nodes_with_entry_count = map
                         .nodes_with_entry_count
                         .checked_sub(1)
                         .expect("nodes_with_entry_count should be >= 0");
                 }
                 if dropped.had_copy_source {
                     map.nodes_with_copy_source_count = map
                         .nodes_with_copy_source_count
                         .checked_sub(1)
                         .expect("nodes_with_copy_source_count should be >= 0");
                 }
             } else {
                 debug_assert!(!was_tracked);
             }
             Ok(())
         })
     }
 
     pub fn has_tracked_dir(
         &mut self,
         directory: &HgPath,
     ) -> Result<bool, DirstateError> {
         self.with_dmap_mut(|map| {
             if let Some(node) = map.get_node(directory)? {
                 // A node without a `DirstateEntry` was created to hold child
                 // nodes, and is therefore a directory.
                 let is_dir = node.entry()?.is_none();
                 Ok(is_dir && node.tracked_descendants_count() > 0)
             } else {
                 Ok(false)
             }
         })
     }
 
     pub fn has_dir(
         &mut self,
         directory: &HgPath,
     ) -> Result<bool, DirstateError> {
         self.with_dmap_mut(|map| {
             if let Some(node) = map.get_node(directory)? {
                 // A node without a `DirstateEntry` was created to hold child
                 // nodes, and is therefore a directory.
                 let is_dir = node.entry()?.is_none();
                 Ok(is_dir && node.descendants_with_entry_count() > 0)
             } else {
                 Ok(false)
             }
         })
     }
 
     #[timed]
     pub fn pack_v1(
         &self,
         parents: DirstateParents,
     ) -> Result<Vec<u8>, DirstateError> {
         let map = self.get_map();
         // Optizimation (to be measured?): pre-compute size to avoid `Vec`
         // reallocations
         let mut size = parents.as_bytes().len();
         for node in map.iter_nodes() {
             let node = node?;
             if node.entry()?.is_some() {
                 size += packed_entry_size(
                     node.full_path(map.on_disk)?,
                     node.copy_source(map.on_disk)?,
                 );
             }
         }
 
         let mut packed = Vec::with_capacity(size);
         packed.extend(parents.as_bytes());
 
         for node in map.iter_nodes() {
             let node = node?;
             if let Some(entry) = node.entry()? {
                 pack_entry(
                     node.full_path(map.on_disk)?,
                     &entry,
                     node.copy_source(map.on_disk)?,
                     &mut packed,
                 );
             }
         }
         Ok(packed)
     }
 
     /// Returns new data and metadata together with whether that data should be
     /// appended to the existing data file whose content is at
     /// `map.on_disk` (true), instead of written to a new data file
     /// (false), and the previous size of data on disk.
     #[timed]
     pub fn pack_v2(
         &self,
         write_mode: DirstateMapWriteMode,
     ) -> Result<(Vec<u8>, on_disk::TreeMetadata, bool, usize), DirstateError>
     {
         let map = self.get_map();
         on_disk::write(map, write_mode)
     }
 
     /// `callback` allows the caller to process and do something with the
     /// results of the status. This is needed to do so efficiently (i.e.
     /// without cloning the `DirstateStatus` object with its paths) because
     /// we need to borrow from `Self`.
     pub fn with_status<R>(
         &mut self,
         matcher: &(dyn Matcher + Sync),
         root_dir: PathBuf,
         ignore_files: Vec<PathBuf>,
         options: StatusOptions,
         callback: impl for<'r> FnOnce(
             Result<(DirstateStatus<'r>, Vec<PatternFileWarning>), StatusError>,
         ) -> R,
     ) -> R {
         self.with_dmap_mut(|map| {
             callback(super::status::status(
                 map,
                 matcher,
                 root_dir,
                 ignore_files,
                 options,
             ))
         })
     }
 
     pub fn copy_map_len(&self) -> usize {
         let map = self.get_map();
         map.nodes_with_copy_source_count as usize
     }
 
     pub fn copy_map_iter(&self) -> CopyMapIter<'_> {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             Ok(if let Some(source) = node.copy_source(map.on_disk)? {
                 Some((node.full_path(map.on_disk)?, source))
             } else {
                 None
             })
         }))
     }
 
     pub fn copy_map_contains_key(
         &self,
         key: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         let map = self.get_map();
         Ok(if let Some(node) = map.get_node(key)? {
             node.has_copy_source()
         } else {
             false
         })
     }
 
     pub fn copy_map_get(
         &self,
         key: &HgPath,
     ) -> Result<Option<&HgPath>, DirstateV2ParseError> {
         let map = self.get_map();
         if let Some(node) = map.get_node(key)? {
             if let Some(source) = node.copy_source(map.on_disk)? {
                 return Ok(Some(source));
             }
         }
         Ok(None)
     }
 
     pub fn copy_map_remove(
         &mut self,
         key: &HgPath,
     ) -> Result<Option<HgPathBuf>, DirstateV2ParseError> {
         self.with_dmap_mut(|map| {
             let count = &mut map.nodes_with_copy_source_count;
             let unreachable_bytes = &mut map.unreachable_bytes;
             Ok(DirstateMap::get_node_mut_inner(
                 map.on_disk,
                 unreachable_bytes,
                 &mut map.root,
                 key,
                 |_ancestor| {},
             )?
             .and_then(|node| {
                 if let Some(source) = &node.copy_source {
                     *count = count
                         .checked_sub(1)
                         .expect("nodes_with_copy_source_count should be >= 0");
                     DirstateMap::count_dropped_path(unreachable_bytes, source);
                 }
                 node.copy_source.take().map(Cow::into_owned)
             }))
         })
     }
 
     pub fn copy_map_insert(
         &mut self,
         key: &HgPath,
         value: &HgPath,
     ) -> Result<Option<HgPathBuf>, DirstateV2ParseError> {
         self.with_dmap_mut(|map| {
             let node = map.get_or_insert_node(&key, |_ancestor| {})?;
             let had_copy_source = node.copy_source.is_none();
             let old = node
                 .copy_source
                 .replace(value.to_owned().into())
                 .map(Cow::into_owned);
             if had_copy_source {
                 map.nodes_with_copy_source_count += 1
             }
             Ok(old)
         })
     }
 
     pub fn len(&self) -> usize {
         let map = self.get_map();
         map.nodes_with_entry_count as usize
     }
 
     pub fn contains_key(
         &self,
         key: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         Ok(self.get(key)?.is_some())
     }
 
     pub fn get(
         &self,
         key: &HgPath,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         let map = self.get_map();
         Ok(if let Some(node) = map.get_node(key)? {
             node.entry()?
         } else {
             None
         })
     }
 
     pub fn iter(&self) -> StateMapIter<'_> {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             Ok(if let Some(entry) = node.entry()? {
                 Some((node.full_path(map.on_disk)?, entry))
             } else {
                 None
             })
         }))
     }
 
     pub fn iter_tracked_dirs(
         &mut self,
     ) -> Result<
         Box<
             dyn Iterator<Item = Result<&HgPath, DirstateV2ParseError>>
                 + Send
                 + '_,
         >,
         DirstateError,
     > {
         let map = self.get_map();
         let on_disk = map.on_disk;
         Ok(Box::new(filter_map_results(
             map.iter_nodes(),
             move |node| {
                 Ok(if node.tracked_descendants_count() > 0 {
                     Some(node.full_path(on_disk)?)
                 } else {
                     None
                 })
             },
         )))
     }
 
     /// Only public because it needs to be exposed to the Python layer.
     /// It is not the full `setparents` logic, only the parts that mutate the
     /// entries.
     pub fn setparents_fixup(
         &mut self,
     ) -> Result<Vec<(HgPathBuf, HgPathBuf)>, DirstateV2ParseError> {
         // XXX
         // All the copying and re-querying is quite inefficient, but this is
         // still a lot better than doing it from Python.
         //
         // The better solution is to develop a mechanism for `iter_mut`,
         // which will be a lot more involved: we're dealing with a lazy,
         // append-mostly, tree-like data structure. This will do for now.
         let mut copies = vec![];
         let mut files_with_p2_info = vec![];
         for res in self.iter() {
             let (path, entry) = res?;
             if entry.p2_info() {
                 files_with_p2_info.push(path.to_owned())
             }
         }
         self.with_dmap_mut(|map| {
             for path in files_with_p2_info.iter() {
                 let node = map.get_or_insert_node(path, |_| {})?;
                 let entry =
                     node.data.as_entry_mut().expect("entry should exist");
                 entry.drop_merge_data();
                 if let Some(source) = node.copy_source.take().as_deref() {
                     copies.push((path.to_owned(), source.to_owned()));
                 }
             }
             Ok(copies)
         })
     }
 
     pub fn debug_iter(
         &self,
         all: bool,
     ) -> Box<
         dyn Iterator<
                 Item = Result<
                     (&HgPath, (u8, i32, i32, i32)),
                     DirstateV2ParseError,
                 >,
             > + Send
             + '_,
     > {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             let debug_tuple = if let Some(entry) = node.entry()? {
                 entry.debug_tuple()
             } else if !all {
                 return Ok(None);
             } else if let Some(mtime) = node.cached_directory_mtime()? {
                 (b' ', 0, -1, mtime.truncated_seconds() as i32)
             } else {
                 (b' ', 0, -1, -1)
             };
             Ok(Some((node.full_path(map.on_disk)?, debug_tuple)))
         }))
     }
 }
 #[cfg(test)]
 mod tests {
     use super::*;
 
     /// Shortcut to return tracked descendants of a path.
     /// Panics if the path does not exist.
     fn tracked_descendants(map: &OwningDirstateMap, path: &[u8]) -> u32 {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         node.unwrap().unwrap().tracked_descendants_count()
     }
 
     /// Shortcut to return descendants with an entry.
     /// Panics if the path does not exist.
     fn descendants_with_an_entry(map: &OwningDirstateMap, path: &[u8]) -> u32 {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         node.unwrap().unwrap().descendants_with_entry_count()
     }
 
     fn assert_does_not_exist(map: &OwningDirstateMap, path: &[u8]) {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         assert!(node.unwrap().is_none());
     }
 
     /// Shortcut for path creation in tests
     fn p(b: &[u8]) -> &HgPath {
         HgPath::new(b)
     }
 
     /// Test the very simple case a single tracked file
     #[test]
     fn test_tracked_descendants_simple() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
         assert_eq!(map.len(), 0);
 
         map.set_tracked(p(b"some/nested/path"))?;
 
         assert_eq!(map.len(), 1);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested/path"), 0);
 
         map.set_untracked(p(b"some/nested/path"))?;
         assert_eq!(map.len(), 0);
         assert!(map.get_map().get_node(p(b"some"))?.is_none());
 
         Ok(())
     }
 
     /// Test the simple case of all tracked, but multiple files
     #[test]
     fn test_tracked_descendants_multiple() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         map.set_tracked(p(b"some/nested/path"))?;
         map.set_tracked(p(b"some/nested/file"))?;
         // one layer without any files to test deletion cascade
         map.set_tracked(p(b"some/other/nested/path"))?;
         map.set_tracked(p(b"root_file"))?;
         map.set_tracked(p(b"some/file"))?;
         map.set_tracked(p(b"some/file2"))?;
         map.set_tracked(p(b"some/file3"))?;
 
         assert_eq!(map.len(), 7);
         assert_eq!(tracked_descendants(&map, b"some"), 6);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 2);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested/path"), 0);
 
         map.set_untracked(p(b"some/nested/path"))?;
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 5);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
 
         map.set_untracked(p(b"some/nested/file"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 4);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_does_not_exist(&map, b"some_nested");
 
         map.set_untracked(p(b"some/other/nested/path"))?;
         assert_eq!(map.len(), 4);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_does_not_exist(&map, b"some/other");
 
         map.set_untracked(p(b"root_file"))?;
         assert_eq!(map.len(), 3);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_does_not_exist(&map, b"root_file");
 
         map.set_untracked(p(b"some/file"))?;
         assert_eq!(map.len(), 2);
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_does_not_exist(&map, b"some/file");
 
         map.set_untracked(p(b"some/file2"))?;
         assert_eq!(map.len(), 1);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_does_not_exist(&map, b"some/file2");
 
         map.set_untracked(p(b"some/file3"))?;
         assert_eq!(map.len(), 0);
         assert_does_not_exist(&map, b"some/file3");
 
         Ok(())
     }
 
     /// Check with a mix of tracked and non-tracked items
     #[test]
     fn test_tracked_descendants_different() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // A file that was just added
         map.set_tracked(p(b"some/nested/path"))?;
         // This has no information, the dirstate should ignore it
         map.reset_state(p(b"some/file"), false, false, false, false, None)?;
         assert_does_not_exist(&map, b"some/file");
 
         // A file that was removed
         map.reset_state(
             p(b"some/nested/file"),
             false,
             true,
             false,
             false,
             None,
         )?;
         assert!(!map.get(p(b"some/nested/file"))?.unwrap().tracked());
         // Only present in p2
         map.reset_state(p(b"some/file3"), false, false, true, false, None)?;
         assert!(!map.get(p(b"some/file3"))?.unwrap().tracked());
         // A file that was merged
         map.reset_state(p(b"root_file"), true, true, true, false, None)?;
         assert!(map.get(p(b"root_file"))?.unwrap().tracked());
         // A file that is added, with info from p2
         // XXX is that actually possible?
         map.reset_state(p(b"some/file2"), true, false, true, false, None)?;
         assert!(map.get(p(b"some/file2"))?.unwrap().tracked());
         // A clean file
         // One layer without any files to test deletion cascade
         map.reset_state(
             p(b"some/other/nested/path"),
             true,
             true,
             false,
             false,
             None,
         )?;
         assert!(map.get(p(b"some/other/nested/path"))?.unwrap().tracked());
 
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 5);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested/path"), 0);
         assert_eq!(
             descendants_with_an_entry(&map, b"some/other/nested/path"),
             0
         );
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
 
         // might as well check this
         map.set_untracked(p(b"path/does/not/exist"))?;
         assert_eq!(map.len(), 6);
 
         map.set_untracked(p(b"some/other/nested/path"))?;
         // It is set untracked but not deleted since it held other information
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 5);
         assert_eq!(descendants_with_an_entry(&map, b"some/other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
 
         map.set_untracked(p(b"some/nested/path"))?;
         // It is set untracked *and* deleted since it was only added
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 4);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 0);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 1);
         assert_does_not_exist(&map, b"some/nested/path");
 
         map.set_untracked(p(b"root_file"))?;
         // Untracked but not deleted
         assert_eq!(map.len(), 5);
         assert!(map.get(p(b"root_file"))?.is_some());
 
         map.set_untracked(p(b"some/file2"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 0);
         assert!(map.get(p(b"some/file2"))?.is_some());
 
         map.set_untracked(p(b"some/file3"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 0);
         assert!(map.get(p(b"some/file3"))?.is_some());
 
         Ok(())
     }
 
     /// Check that copies counter is correctly updated
     #[test]
     fn test_copy_source() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // Clean file
         map.reset_state(p(b"files/clean"), true, true, false, false, None)?;
         // Merged file
         map.reset_state(p(b"files/from_p2"), true, true, true, false, None)?;
         // Removed file
         map.reset_state(p(b"removed"), false, true, false, false, None)?;
         // Added file
         map.reset_state(p(b"files/added"), true, false, false, false, None)?;
         // Add copy
         map.copy_map_insert(p(b"files/clean"), p(b"clean_copy_source"))?;
         assert_eq!(map.copy_map_len(), 1);
 
         // Copy override
         map.copy_map_insert(p(b"files/clean"), p(b"other_clean_copy_source"))?;
         assert_eq!(map.copy_map_len(), 1);
 
         // Multiple copies
         map.copy_map_insert(p(b"removed"), p(b"removed_copy_source"))?;
         assert_eq!(map.copy_map_len(), 2);
 
         map.copy_map_insert(p(b"files/added"), p(b"added_copy_source"))?;
         assert_eq!(map.copy_map_len(), 3);
 
         // Added, so the entry is completely removed
         map.set_untracked(p(b"files/added"))?;
         assert_does_not_exist(&map, b"files/added");
         assert_eq!(map.copy_map_len(), 2);
 
         // Removed, so the entry is kept around, so is its copy
         map.set_untracked(p(b"removed"))?;
         assert!(map.get(p(b"removed"))?.is_some());
         assert_eq!(map.copy_map_len(), 2);
 
         // Clean, so the entry is kept around, but not its copy
         map.set_untracked(p(b"files/clean"))?;
         assert!(map.get(p(b"files/clean"))?.is_some());
         assert_eq!(map.copy_map_len(), 1);
 
         map.copy_map_insert(p(b"files/from_p2"), p(b"from_p2_copy_source"))?;
         assert_eq!(map.copy_map_len(), 2);
 
         // Info from p2, so its copy source info is kept around
         map.set_untracked(p(b"files/from_p2"))?;
         assert!(map.get(p(b"files/from_p2"))?.is_some());
         assert_eq!(map.copy_map_len(), 2);
 
         Ok(())
     }
 
     /// Test with "on disk" data. For the sake of this test, the "on disk" data
     /// does not actually come from the disk, but it's opaque to the code being
     /// tested.
     #[test]
     fn test_on_disk() -> Result<(), DirstateError> {
         // First let's create some data to put "on disk"
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // A file that was just added
         map.set_tracked(p(b"some/nested/added"))?;
         map.copy_map_insert(p(b"some/nested/added"), p(b"added_copy_source"))?;
 
         // A file that was removed
         map.reset_state(
             p(b"some/nested/removed"),
             false,
             true,
             false,
             false,
             None,
         )?;
         // Only present in p2
         map.reset_state(
             p(b"other/p2_info_only"),
             false,
             false,
             true,
             false,
             None,
         )?;
         map.copy_map_insert(
             p(b"other/p2_info_only"),
             p(b"other/p2_info_copy_source"),
         )?;
         // A file that was merged
         map.reset_state(p(b"merged"), true, true, true, false, None)?;
         // A file that is added, with info from p2
         // XXX is that actually possible?
         map.reset_state(
             p(b"other/added_with_p2"),
             true,
             false,
             true,
             false,
             None,
         )?;
         // One layer without any files to test deletion cascade
         // A clean file
         map.reset_state(
             p(b"some/other/nested/clean"),
             true,
             true,
             false,
             false,
             None,
         )?;
 
         let (packed, metadata, _should_append, _old_data_size) =
             map.pack_v2(DirstateMapWriteMode::ForceNewDataFile)?;
         let packed_len = packed.len();
         assert!(packed_len > 0);
 
         // Recreate "from disk"
         let mut map = OwningDirstateMap::new_v2(
             packed,
             packed_len,
             metadata.as_bytes(),
         )?;
 
         // Check that everything is accounted for
         assert!(map.contains_key(p(b"some/nested/added"))?);
         assert!(map.contains_key(p(b"some/nested/removed"))?);
         assert!(map.contains_key(p(b"merged"))?);
         assert!(map.contains_key(p(b"other/p2_info_only"))?);
         assert!(map.contains_key(p(b"other/added_with_p2"))?);
         assert!(map.contains_key(p(b"some/other/nested/clean"))?);
         assert_eq!(
             map.copy_map_get(p(b"some/nested/added"))?,
             Some(p(b"added_copy_source"))
         );
         assert_eq!(
             map.copy_map_get(p(b"other/p2_info_only"))?,
             Some(p(b"other/p2_info_copy_source"))
         );
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 3);
         assert_eq!(tracked_descendants(&map, b"other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"other"), 2);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
         assert_eq!(map.len(), 6);
         assert_eq!(map.get_map().unreachable_bytes, 0);
         assert_eq!(map.copy_map_len(), 2);
 
         // Shouldn't change anything since it's already not tracked
         map.set_untracked(p(b"some/nested/removed"))?;
         assert_eq!(map.get_map().unreachable_bytes, 0);
 
         match map.get_map().root {
             ChildNodes::InMemory(_) => {
                 panic!("root should not have been mutated")
             }
             _ => (),
         }
         // We haven't mutated enough (nothing, actually), we should still be in
         // the append strategy
         assert!(map.get_map().write_should_append());
 
         // But this mutates the structure, so there should be unreachable_bytes
         assert!(map.set_untracked(p(b"some/nested/added"))?);
         let unreachable_bytes = map.get_map().unreachable_bytes;
         assert!(unreachable_bytes > 0);
 
         match map.get_map().root {
             ChildNodes::OnDisk(_) => panic!("root should have been mutated"),
             _ => (),
         }
 
         // This should not mutate the structure either, since `root` has
         // already been mutated along with its direct children.
         map.set_untracked(p(b"merged"))?;
         assert_eq!(map.get_map().unreachable_bytes, unreachable_bytes);
 
         match map.get_map().get_node(p(b"other/added_with_p2"))?.unwrap() {
             NodeRef::InMemory(_, _) => {
                 panic!("'other/added_with_p2' should not have been mutated")
             }
             _ => (),
         }
         // But this should, since it's in a different path
         // than `<root>some/nested/add`
         map.set_untracked(p(b"other/added_with_p2"))?;
         assert!(map.get_map().unreachable_bytes > unreachable_bytes);
 
         match map.get_map().get_node(p(b"other/added_with_p2"))?.unwrap() {
             NodeRef::OnDisk(_) => {
                 panic!("'other/added_with_p2' should have been mutated")
             }
             _ => (),
         }
 
         // We have rewritten most of the tree, we should create a new file
         assert!(!map.get_map().write_should_append());
 
         Ok(())
     }
 }

rust/hg-core/src/dirstate_tree/on_disk.rs

@@ -1,888 +1,890 b''
 //! The "version 2" disk representation of the dirstate
 //!
 //! See `mercurial/helptext/internals/dirstate-v2.txt`
 
 use crate::dirstate::{DirstateV2Data, TruncatedTimestamp};
 use crate::dirstate_tree::dirstate_map::DirstateVersion;
 use crate::dirstate_tree::dirstate_map::{
     self, DirstateMap, DirstateMapWriteMode, NodeRef,
 };
 use crate::dirstate_tree::path_with_basename::WithBasename;
 use crate::errors::HgError;
 use crate::utils::hg_path::HgPath;
 use crate::DirstateEntry;
 use crate::DirstateError;
 use crate::DirstateParents;
 use bitflags::bitflags;
 use bytes_cast::unaligned::{U16Be, U32Be};
 use bytes_cast::BytesCast;
 use format_bytes::format_bytes;
 use rand::Rng;
 use std::borrow::Cow;
 use std::convert::{TryFrom, TryInto};
 use std::fmt::Write;
 
 /// Added at the start of `.hg/dirstate` when the "v2" format is used.
 /// This a redundant sanity check more than an actual "magic number" since
 /// `.hg/requires` already governs which format should be used.
 pub const V2_FORMAT_MARKER: &[u8; 12] = b"dirstate-v2\n";
 
 /// Keep space for 256-bit hashes
 const STORED_NODE_ID_BYTES: usize = 32;
 
 /// … even though only 160 bits are used for now, with SHA-1
 const USED_NODE_ID_BYTES: usize = 20;
 
 pub(super) const IGNORE_PATTERNS_HASH_LEN: usize = 20;
 pub(super) type IgnorePatternsHash = [u8; IGNORE_PATTERNS_HASH_LEN];
 
 /// Must match constants of the same names in `mercurial/dirstateutils/v2.py`
 const TREE_METADATA_SIZE: usize = 44;
 const NODE_SIZE: usize = 44;
 
 /// Make sure that size-affecting changes are made knowingly
 #[allow(unused)]
 fn static_assert_size_of() {
     let _ = std::mem::transmute::<TreeMetadata, [u8; TREE_METADATA_SIZE]>;
     let _ = std::mem::transmute::<DocketHeader, [u8; TREE_METADATA_SIZE + 81]>;
     let _ = std::mem::transmute::<Node, [u8; NODE_SIZE]>;
 }
 
 // Must match `HEADER` in `mercurial/dirstateutils/docket.py`
 #[derive(BytesCast)]
 #[repr(C)]
 struct DocketHeader {
     marker: [u8; V2_FORMAT_MARKER.len()],
     parent_1: [u8; STORED_NODE_ID_BYTES],
     parent_2: [u8; STORED_NODE_ID_BYTES],
 
     metadata: TreeMetadata,
 
     /// Counted in bytes
     data_size: Size,
 
     uuid_size: u8,
 }
 
 pub struct Docket<'on_disk> {
     header: &'on_disk DocketHeader,
     pub uuid: &'on_disk [u8],
 }
 
 /// Fields are documented in the *Tree metadata in the docket file*
 /// section of `mercurial/helptext/internals/dirstate-v2.txt`
 #[derive(BytesCast)]
 #[repr(C)]
 pub struct TreeMetadata {
     root_nodes: ChildNodes,
     nodes_with_entry_count: Size,
     nodes_with_copy_source_count: Size,
     unreachable_bytes: Size,
     unused: [u8; 4],
 
     /// See *Optional hash of ignore patterns* section of
     /// `mercurial/helptext/internals/dirstate-v2.txt`
     ignore_patterns_hash: IgnorePatternsHash,
 }
 
 /// Fields are documented in the *The data file format*
 /// section of `mercurial/helptext/internals/dirstate-v2.txt`
 #[derive(BytesCast, Debug)]
 #[repr(C)]
 pub(super) struct Node {
     full_path: PathSlice,
 
     /// In bytes from `self.full_path.start`
     base_name_start: PathSize,
 
     copy_source: OptPathSlice,
     children: ChildNodes,
     pub(super) descendants_with_entry_count: Size,
     pub(super) tracked_descendants_count: Size,
     flags: U16Be,
     size: U32Be,
     mtime: PackedTruncatedTimestamp,
 }
 
 bitflags! {
     #[repr(C)]
     struct Flags: u16 {
         const WDIR_TRACKED = 1 << 0;
         const P1_TRACKED = 1 << 1;
         const P2_INFO = 1 << 2;
         const MODE_EXEC_PERM = 1 << 3;
         const MODE_IS_SYMLINK = 1 << 4;
         const HAS_FALLBACK_EXEC = 1 <<  5;
         const FALLBACK_EXEC = 1 <<  6;
         const HAS_FALLBACK_SYMLINK = 1 <<  7;
         const FALLBACK_SYMLINK = 1 <<  8;
         const EXPECTED_STATE_IS_MODIFIED = 1 << 9;
         const HAS_MODE_AND_SIZE = 1 <<10;
         const HAS_MTIME = 1 <<11;
         const MTIME_SECOND_AMBIGUOUS = 1 << 12;
         const DIRECTORY = 1 <<13;
         const ALL_UNKNOWN_RECORDED = 1 <<14;
         const ALL_IGNORED_RECORDED = 1 <<15;
     }
 }
 
 /// Duration since the Unix epoch
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct PackedTruncatedTimestamp {
     truncated_seconds: U32Be,
     nanoseconds: U32Be,
 }
 
 /// Counted in bytes from the start of the file
 ///
 /// NOTE: not supporting `.hg/dirstate` files larger than 4 GiB.
 type Offset = U32Be;
 
 /// Counted in number of items
 ///
 /// NOTE: we choose not to support counting more than 4 billion nodes anywhere.
 type Size = U32Be;
 
 /// Counted in bytes
 ///
 /// NOTE: we choose not to support file names/paths longer than 64 KiB.
 type PathSize = U16Be;
 
 /// A contiguous sequence of `len` times `Node`, representing the child nodes
 /// of either some other node or of the repository root.
 ///
 /// Always sorted by ascending `full_path`, to allow binary search.
 /// Since nodes with the same parent nodes also have the same parent path,
 /// only the `base_name`s need to be compared during binary search.
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct ChildNodes {
     start: Offset,
     len: Size,
 }
 
 /// A `HgPath` of `len` bytes
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct PathSlice {
     start: Offset,
     len: PathSize,
 }
 
 /// Either nothing if `start == 0`, or a `HgPath` of `len` bytes
 type OptPathSlice = PathSlice;
 
 /// Unexpected file format found in `.hg/dirstate` with the "v2" format.
 ///
 /// This should only happen if Mercurial is buggy or a repository is corrupted.
 #[derive(Debug)]
 pub struct DirstateV2ParseError {
     message: String,
 }
 
 impl DirstateV2ParseError {
     pub fn new<S: Into<String>>(message: S) -> Self {
         Self {
             message: message.into(),
         }
     }
 }
 
 impl From<DirstateV2ParseError> for HgError {
     fn from(e: DirstateV2ParseError) -> Self {
         HgError::corrupted(format!("dirstate-v2 parse error: {}", e.message))
     }
 }
 
 impl From<DirstateV2ParseError> for crate::DirstateError {
     fn from(error: DirstateV2ParseError) -> Self {
         HgError::from(error).into()
     }
 }
 
 impl TreeMetadata {
     pub fn as_bytes(&self) -> &[u8] {
         BytesCast::as_bytes(self)
     }
 }
 
 impl<'on_disk> Docket<'on_disk> {
     /// Generate the identifier for a new data file
     ///
     /// TODO: support the `HGTEST_UUIDFILE` environment variable.
     /// See `mercurial/revlogutils/docket.py`
     pub fn new_uid() -> String {
         const ID_LENGTH: usize = 8;
         let mut id = String::with_capacity(ID_LENGTH);
         let mut rng = rand::thread_rng();
         for _ in 0..ID_LENGTH {
             // One random hexadecimal digit.
             // `unwrap` never panics because `impl Write for String`
             // never returns an error.
             write!(&mut id, "{:x}", rng.gen_range(0..16)).unwrap();
         }
         id
     }
 
     pub fn serialize(
         parents: DirstateParents,
         tree_metadata: TreeMetadata,
         data_size: u64,
         uuid: &[u8],
     ) -> Result<Vec<u8>, std::num::TryFromIntError> {
         let header = DocketHeader {
             marker: *V2_FORMAT_MARKER,
             parent_1: parents.p1.pad_to_256_bits(),
             parent_2: parents.p2.pad_to_256_bits(),
             metadata: tree_metadata,
             data_size: u32::try_from(data_size)?.into(),
             uuid_size: uuid.len().try_into()?,
         };
         let header = header.as_bytes();
         let mut docket = Vec::with_capacity(header.len() + uuid.len());
         docket.extend_from_slice(header);
         docket.extend_from_slice(uuid);
         Ok(docket)
     }
 
     pub fn parents(&self) -> DirstateParents {
         use crate::Node;
         let p1 = Node::try_from(&self.header.parent_1[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         let p2 = Node::try_from(&self.header.parent_2[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         DirstateParents { p1, p2 }
     }
 
     pub fn tree_metadata(&self) -> &[u8] {
         self.header.metadata.as_bytes()
     }
 
     pub fn data_size(&self) -> usize {
         // This `unwrap` could only panic on a 16-bit CPU
         self.header.data_size.get().try_into().unwrap()
     }
 
     pub fn data_filename(&self) -> String {
         String::from_utf8(format_bytes!(b"dirstate.{}", self.uuid)).unwrap()
     }
 }
 
 pub fn read_docket(
     on_disk: &[u8],
 ) -> Result<Docket<'_>, DirstateV2ParseError> {
     let (header, uuid) = DocketHeader::from_bytes(on_disk).map_err(|e| {
         DirstateV2ParseError::new(format!("when reading docket, {}", e))
     })?;
     let uuid_size = header.uuid_size as usize;
     if header.marker == *V2_FORMAT_MARKER && uuid.len() == uuid_size {
         Ok(Docket { header, uuid })
     } else {
         Err(DirstateV2ParseError::new(
             "invalid format marker or uuid size",
         ))
     }
 }
 
 pub(super) fn read<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
 ) -> Result<DirstateMap<'on_disk>, DirstateV2ParseError> {
     if on_disk.is_empty() {
         let mut map = DirstateMap::empty(on_disk);
         map.dirstate_version = DirstateVersion::V2;
         return Ok(map);
     }
     let (meta, _) = TreeMetadata::from_bytes(metadata).map_err(|e| {
         DirstateV2ParseError::new(format!("when parsing tree metadata, {}", e))
     })?;
     let dirstate_map = DirstateMap {
         on_disk,
         root: dirstate_map::ChildNodes::OnDisk(
             read_nodes(on_disk, meta.root_nodes).map_err(|mut e| {
                 e.message = format!("{}, when reading root notes", e.message);
                 e
             })?,
         ),
         nodes_with_entry_count: meta.nodes_with_entry_count.get(),
         nodes_with_copy_source_count: meta.nodes_with_copy_source_count.get(),
         ignore_patterns_hash: meta.ignore_patterns_hash,
         unreachable_bytes: meta.unreachable_bytes.get(),
         old_data_size: on_disk.len(),
         dirstate_version: DirstateVersion::V2,
+        write_mode: DirstateMapWriteMode::Auto,
     };
     Ok(dirstate_map)
 }
 
 impl Node {
     pub(super) fn full_path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         read_hg_path(on_disk, self.full_path)
     }
 
     pub(super) fn base_name_start<'on_disk>(
         &self,
     ) -> Result<usize, DirstateV2ParseError> {
         let start = self.base_name_start.get();
         if start < self.full_path.len.get() {
             let start = usize::try_from(start)
                 // u32 -> usize, could only panic on a 16-bit CPU
                 .expect("dirstate-v2 base_name_start out of bounds");
             Ok(start)
         } else {
             Err(DirstateV2ParseError::new("not enough bytes for base name"))
         }
     }
 
     pub(super) fn base_name<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         let full_path = self.full_path(on_disk)?;
         let base_name_start = self.base_name_start()?;
         Ok(HgPath::new(&full_path.as_bytes()[base_name_start..]))
     }
 
     pub(super) fn path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::NodeKey<'on_disk>, DirstateV2ParseError> {
         Ok(WithBasename::from_raw_parts(
             Cow::Borrowed(self.full_path(on_disk)?),
             self.base_name_start()?,
         ))
     }
 
     pub(super) fn has_copy_source<'on_disk>(&self) -> bool {
         self.copy_source.start.get() != 0
     }
 
     pub(super) fn copy_source<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<&'on_disk HgPath>, DirstateV2ParseError> {
         Ok(if self.has_copy_source() {
             Some(read_hg_path(on_disk, self.copy_source)?)
         } else {
             None
         })
     }
 
     fn flags(&self) -> Flags {
         Flags::from_bits_truncate(self.flags.get())
     }
 
     fn has_entry(&self) -> bool {
         self.flags().intersects(
             Flags::WDIR_TRACKED | Flags::P1_TRACKED | Flags::P2_INFO,
         )
     }
 
     pub(super) fn node_data(
         &self,
     ) -> Result<dirstate_map::NodeData, DirstateV2ParseError> {
         if self.has_entry() {
             Ok(dirstate_map::NodeData::Entry(self.assume_entry()?))
         } else if let Some(mtime) = self.cached_directory_mtime()? {
             Ok(dirstate_map::NodeData::CachedDirectory { mtime })
         } else {
             Ok(dirstate_map::NodeData::None)
         }
     }
 
     pub(super) fn cached_directory_mtime(
         &self,
     ) -> Result<Option<TruncatedTimestamp>, DirstateV2ParseError> {
         // For now we do not have code to handle the absence of
         // ALL_UNKNOWN_RECORDED, so we ignore the mtime if the flag is
         // unset.
         if self.flags().contains(Flags::DIRECTORY)
             && self.flags().contains(Flags::HAS_MTIME)
             && self.flags().contains(Flags::ALL_UNKNOWN_RECORDED)
         {
             Ok(Some(self.mtime()?))
         } else {
             Ok(None)
         }
     }
 
     fn synthesize_unix_mode(&self) -> u32 {
         let file_type = if self.flags().contains(Flags::MODE_IS_SYMLINK) {
             libc::S_IFLNK
         } else {
             libc::S_IFREG
         };
         let permisions = if self.flags().contains(Flags::MODE_EXEC_PERM) {
             0o755
         } else {
             0o644
         };
         (file_type | permisions).into()
     }
 
     fn mtime(&self) -> Result<TruncatedTimestamp, DirstateV2ParseError> {
         let mut m: TruncatedTimestamp = self.mtime.try_into()?;
         if self.flags().contains(Flags::MTIME_SECOND_AMBIGUOUS) {
             m.second_ambiguous = true;
         }
         Ok(m)
     }
 
     fn assume_entry(&self) -> Result<DirstateEntry, DirstateV2ParseError> {
         // TODO: convert through raw bits instead?
         let wc_tracked = self.flags().contains(Flags::WDIR_TRACKED);
         let p1_tracked = self.flags().contains(Flags::P1_TRACKED);
         let p2_info = self.flags().contains(Flags::P2_INFO);
         let mode_size = if self.flags().contains(Flags::HAS_MODE_AND_SIZE)
             && !self.flags().contains(Flags::EXPECTED_STATE_IS_MODIFIED)
         {
             Some((self.synthesize_unix_mode(), self.size.into()))
         } else {
             None
         };
         let mtime = if self.flags().contains(Flags::HAS_MTIME)
             && !self.flags().contains(Flags::DIRECTORY)
             && !self.flags().contains(Flags::EXPECTED_STATE_IS_MODIFIED)
         {
             Some(self.mtime()?)
         } else {
             None
         };
         let fallback_exec = if self.flags().contains(Flags::HAS_FALLBACK_EXEC)
         {
             Some(self.flags().contains(Flags::FALLBACK_EXEC))
         } else {
             None
         };
         let fallback_symlink =
             if self.flags().contains(Flags::HAS_FALLBACK_SYMLINK) {
                 Some(self.flags().contains(Flags::FALLBACK_SYMLINK))
             } else {
                 None
             };
         Ok(DirstateEntry::from_v2_data(DirstateV2Data {
             wc_tracked,
             p1_tracked,
             p2_info,
             mode_size,
             mtime,
             fallback_exec,
             fallback_symlink,
         }))
     }
 
     pub(super) fn entry(
         &self,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         if self.has_entry() {
             Ok(Some(self.assume_entry()?))
         } else {
             Ok(None)
         }
     }
 
     pub(super) fn children<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk [Node], DirstateV2ParseError> {
         read_nodes(on_disk, self.children)
     }
 
     pub(super) fn to_in_memory_node<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::Node<'on_disk>, DirstateV2ParseError> {
         Ok(dirstate_map::Node {
             children: dirstate_map::ChildNodes::OnDisk(
                 self.children(on_disk)?,
             ),
             copy_source: self.copy_source(on_disk)?.map(Cow::Borrowed),
             data: self.node_data()?,
             descendants_with_entry_count: self
                 .descendants_with_entry_count
                 .get(),
             tracked_descendants_count: self.tracked_descendants_count.get(),
         })
     }
 
     fn from_dirstate_entry(
         entry: &DirstateEntry,
     ) -> (Flags, U32Be, PackedTruncatedTimestamp) {
         let DirstateV2Data {
             wc_tracked,
             p1_tracked,
             p2_info,
             mode_size: mode_size_opt,
             mtime: mtime_opt,
             fallback_exec,
             fallback_symlink,
         } = entry.v2_data();
         // TODO: convert through raw flag bits instead?
         let mut flags = Flags::empty();
         flags.set(Flags::WDIR_TRACKED, wc_tracked);
         flags.set(Flags::P1_TRACKED, p1_tracked);
         flags.set(Flags::P2_INFO, p2_info);
         let size = if let Some((m, s)) = mode_size_opt {
             let exec_perm = m & (libc::S_IXUSR as u32) != 0;
             let is_symlink = m & (libc::S_IFMT as u32) == libc::S_IFLNK as u32;
             flags.set(Flags::MODE_EXEC_PERM, exec_perm);
             flags.set(Flags::MODE_IS_SYMLINK, is_symlink);
             flags.insert(Flags::HAS_MODE_AND_SIZE);
             s.into()
         } else {
             0.into()
         };
         let mtime = if let Some(m) = mtime_opt {
             flags.insert(Flags::HAS_MTIME);
             if m.second_ambiguous {
                 flags.insert(Flags::MTIME_SECOND_AMBIGUOUS);
             };
             m.into()
         } else {
             PackedTruncatedTimestamp::null()
         };
         if let Some(f_exec) = fallback_exec {
             flags.insert(Flags::HAS_FALLBACK_EXEC);
             if f_exec {
                 flags.insert(Flags::FALLBACK_EXEC);
             }
         }
         if let Some(f_symlink) = fallback_symlink {
             flags.insert(Flags::HAS_FALLBACK_SYMLINK);
             if f_symlink {
                 flags.insert(Flags::FALLBACK_SYMLINK);
             }
         }
         (flags, size, mtime)
     }
 }
 
 fn read_hg_path(
     on_disk: &[u8],
     slice: PathSlice,
 ) -> Result<&HgPath, DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get()).map(HgPath::new)
 }
 
 fn read_nodes(
     on_disk: &[u8],
     slice: ChildNodes,
 ) -> Result<&[Node], DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get())
 }
 
 fn read_slice<T, Len>(
     on_disk: &[u8],
     start: Offset,
     len: Len,
 ) -> Result<&[T], DirstateV2ParseError>
 where
     T: BytesCast,
     Len: TryInto<usize>,
 {
     // Either `usize::MAX` would result in "out of bounds" error since a single
     // `&[u8]` cannot occupy the entire addess space.
     let start = start.get().try_into().unwrap_or(std::usize::MAX);
     let len = len.try_into().unwrap_or(std::usize::MAX);
     let bytes = match on_disk.get(start..) {
         Some(bytes) => bytes,
         None => {
             return Err(DirstateV2ParseError::new(
                 "not enough bytes from disk",
             ))
         }
     };
     T::slice_from_bytes(bytes, len)
         .map_err(|e| {
             DirstateV2ParseError::new(format!("when reading a slice, {}", e))
         })
         .map(|(slice, _rest)| slice)
 }
 
 pub(crate) fn for_each_tracked_path<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
     mut f: impl FnMut(&'on_disk HgPath),
 ) -> Result<(), DirstateV2ParseError> {
     let (meta, _) = TreeMetadata::from_bytes(metadata).map_err(|e| {
         DirstateV2ParseError::new(format!("when parsing tree metadata, {}", e))
     })?;
     fn recur<'on_disk>(
         on_disk: &'on_disk [u8],
         nodes: ChildNodes,
         f: &mut impl FnMut(&'on_disk HgPath),
     ) -> Result<(), DirstateV2ParseError> {
         for node in read_nodes(on_disk, nodes)? {
             if let Some(entry) = node.entry()? {
                 if entry.tracked() {
                     f(node.full_path(on_disk)?)
                 }
             }
             recur(on_disk, node.children, f)?
         }
         Ok(())
     }
     recur(on_disk, meta.root_nodes, &mut f)
 }
 
 /// Returns new data and metadata, together with whether that data should be
 /// appended to the existing data file whose content is at
 /// `dirstate_map.on_disk` (true), instead of written to a new data file
 /// (false), and the previous size of data on disk.
 pub(super) fn write(
     dirstate_map: &DirstateMap,
     write_mode: DirstateMapWriteMode,
 ) -> Result<(Vec<u8>, TreeMetadata, bool, usize), DirstateError> {
     let append = match write_mode {
         DirstateMapWriteMode::Auto => dirstate_map.write_should_append(),
         DirstateMapWriteMode::ForceNewDataFile => false,
+        DirstateMapWriteMode::ForceAppend => true,
     };
     if append {
         log::trace!("appending to the dirstate data file");
     } else {
         log::trace!("creating new dirstate data file");
     }
 
     // This ignores the space for paths, and for nodes without an entry.
     // TODO: better estimate? Skip the `Vec` and write to a file directly?
     let size_guess = std::mem::size_of::<Node>()
         * dirstate_map.nodes_with_entry_count as usize;
 
     let mut writer = Writer {
         dirstate_map,
         append,
         out: Vec::with_capacity(size_guess),
     };
 
     let root_nodes = writer.write_nodes(dirstate_map.root.as_ref())?;
 
     let unreachable_bytes = if append {
         dirstate_map.unreachable_bytes
     } else {
         0
     };
     let meta = TreeMetadata {
         root_nodes,
         nodes_with_entry_count: dirstate_map.nodes_with_entry_count.into(),
         nodes_with_copy_source_count: dirstate_map
             .nodes_with_copy_source_count
             .into(),
         unreachable_bytes: unreachable_bytes.into(),
         unused: [0; 4],
         ignore_patterns_hash: dirstate_map.ignore_patterns_hash,
     };
     Ok((writer.out, meta, append, dirstate_map.old_data_size))
 }
 
 struct Writer<'dmap, 'on_disk> {
     dirstate_map: &'dmap DirstateMap<'on_disk>,
     append: bool,
     out: Vec<u8>,
 }
 
 impl Writer<'_, '_> {
     fn write_nodes(
         &mut self,
         nodes: dirstate_map::ChildNodesRef,
     ) -> Result<ChildNodes, DirstateError> {
         // Reuse already-written nodes if possible
         if self.append {
             if let dirstate_map::ChildNodesRef::OnDisk(nodes_slice) = nodes {
                 let start = self.on_disk_offset_of(nodes_slice).expect(
                     "dirstate-v2 OnDisk nodes not found within on_disk",
                 );
                 let len = child_nodes_len_from_usize(nodes_slice.len());
                 return Ok(ChildNodes { start, len });
             }
         }
 
         // `dirstate_map::ChildNodes::InMemory` contains a `HashMap` which has
         // undefined iteration order. Sort to enable binary search in the
         // written file.
         let nodes = nodes.sorted();
         let nodes_len = nodes.len();
 
         // First accumulate serialized nodes in a `Vec`
         let mut on_disk_nodes = Vec::with_capacity(nodes_len);
         for node in nodes {
             let children =
                 self.write_nodes(node.children(self.dirstate_map.on_disk)?)?;
             let full_path = node.full_path(self.dirstate_map.on_disk)?;
             let full_path = self.write_path(full_path.as_bytes());
             let copy_source = if let Some(source) =
                 node.copy_source(self.dirstate_map.on_disk)?
             {
                 self.write_path(source.as_bytes())
             } else {
                 PathSlice {
                     start: 0.into(),
                     len: 0.into(),
                 }
             };
             on_disk_nodes.push(match node {
                 NodeRef::InMemory(path, node) => {
                     let (flags, size, mtime) = match &node.data {
                         dirstate_map::NodeData::Entry(entry) => {
                             Node::from_dirstate_entry(entry)
                         }
                         dirstate_map::NodeData::CachedDirectory { mtime } => {
                             // we currently never set a mtime if unknown file
                             // are present.
                             // So if we have a mtime for a directory, we know
                             // they are no unknown
                             // files and we
                             // blindly set ALL_UNKNOWN_RECORDED.
                             //
                             // We never set ALL_IGNORED_RECORDED since we
                             // don't track that case
                             // currently.
                             let mut flags = Flags::DIRECTORY
                                 | Flags::HAS_MTIME
                                 | Flags::ALL_UNKNOWN_RECORDED;
                             if mtime.second_ambiguous {
                                 flags.insert(Flags::MTIME_SECOND_AMBIGUOUS)
                             }
                             (flags, 0.into(), (*mtime).into())
                         }
                         dirstate_map::NodeData::None => (
                             Flags::DIRECTORY,
                             0.into(),
                             PackedTruncatedTimestamp::null(),
                         ),
                     };
                     Node {
                         children,
                         copy_source,
                         full_path,
                         base_name_start: u16::try_from(path.base_name_start())
                             // Could only panic for paths over 64 KiB
                             .expect("dirstate-v2 path length overflow")
                             .into(),
                         descendants_with_entry_count: node
                             .descendants_with_entry_count
                             .into(),
                         tracked_descendants_count: node
                             .tracked_descendants_count
                             .into(),
                         flags: flags.bits().into(),
                         size,
                         mtime,
                     }
                 }
                 NodeRef::OnDisk(node) => Node {
                     children,
                     copy_source,
                     full_path,
                     ..*node
                 },
             })
         }
         // … so we can write them contiguously, after writing everything else
         // they refer to.
         let start = self.current_offset();
         let len = child_nodes_len_from_usize(nodes_len);
         self.out.extend(on_disk_nodes.as_bytes());
         Ok(ChildNodes { start, len })
     }
 
     /// If the given slice of items is within `on_disk`, returns its offset
     /// from the start of `on_disk`.
     fn on_disk_offset_of<T>(&self, slice: &[T]) -> Option<Offset>
     where
         T: BytesCast,
     {
         fn address_range(slice: &[u8]) -> std::ops::RangeInclusive<usize> {
             let start = slice.as_ptr() as usize;
             let end = start + slice.len();
             start..=end
         }
         let slice_addresses = address_range(slice.as_bytes());
         let on_disk_addresses = address_range(self.dirstate_map.on_disk);
         if on_disk_addresses.contains(slice_addresses.start())
             && on_disk_addresses.contains(slice_addresses.end())
         {
             let offset = slice_addresses.start() - on_disk_addresses.start();
             Some(offset_from_usize(offset))
         } else {
             None
         }
     }
 
     fn current_offset(&mut self) -> Offset {
         let mut offset = self.out.len();
         if self.append {
             offset += self.dirstate_map.on_disk.len()
         }
         offset_from_usize(offset)
     }
 
     fn write_path(&mut self, slice: &[u8]) -> PathSlice {
         let len = path_len_from_usize(slice.len());
         // Reuse an already-written path if possible
         if self.append {
             if let Some(start) = self.on_disk_offset_of(slice) {
                 return PathSlice { start, len };
             }
         }
         let start = self.current_offset();
         self.out.extend(slice.as_bytes());
         PathSlice { start, len }
     }
 }
 
 fn offset_from_usize(x: usize) -> Offset {
     u32::try_from(x)
         // Could only panic for a dirstate file larger than 4 GiB
         .expect("dirstate-v2 offset overflow")
         .into()
 }
 
 fn child_nodes_len_from_usize(x: usize) -> Size {
     u32::try_from(x)
         // Could only panic with over 4 billion nodes
         .expect("dirstate-v2 slice length overflow")
         .into()
 }
 
 fn path_len_from_usize(x: usize) -> PathSize {
     u16::try_from(x)
         // Could only panic for paths over 64 KiB
         .expect("dirstate-v2 path length overflow")
         .into()
 }
 
 impl From<TruncatedTimestamp> for PackedTruncatedTimestamp {
     fn from(timestamp: TruncatedTimestamp) -> Self {
         Self {
             truncated_seconds: timestamp.truncated_seconds().into(),
             nanoseconds: timestamp.nanoseconds().into(),
         }
     }
 }
 
 impl TryFrom<PackedTruncatedTimestamp> for TruncatedTimestamp {
     type Error = DirstateV2ParseError;
 
     fn try_from(
         timestamp: PackedTruncatedTimestamp,
     ) -> Result<Self, Self::Error> {
         Self::from_already_truncated(
             timestamp.truncated_seconds.get(),
             timestamp.nanoseconds.get(),
             false,
         )
     }
 }
 impl PackedTruncatedTimestamp {
     fn null() -> Self {
         Self {
             truncated_seconds: 0.into(),
             nanoseconds: 0.into(),
         }
     }
 }

rust/hg-core/src/repo.rs

@@ -1,582 +1,599 b''
 use crate::changelog::Changelog;
 use crate::config::{Config, ConfigError, ConfigParseError};
 use crate::dirstate::DirstateParents;
 use crate::dirstate_tree::dirstate_map::DirstateMapWriteMode;
 use crate::dirstate_tree::on_disk::Docket as DirstateDocket;
 use crate::dirstate_tree::owning::OwningDirstateMap;
 use crate::errors::HgResultExt;
 use crate::errors::{HgError, IoResultExt};
 use crate::lock::{try_with_lock_no_wait, LockError};
 use crate::manifest::{Manifest, Manifestlog};
 use crate::revlog::filelog::Filelog;
 use crate::revlog::revlog::RevlogError;
 use crate::utils::files::get_path_from_bytes;
 use crate::utils::hg_path::HgPath;
 use crate::utils::SliceExt;
 use crate::vfs::{is_dir, is_file, Vfs};
 use crate::{requirements, NodePrefix};
 use crate::{DirstateError, Revision};
 use std::cell::{Ref, RefCell, RefMut};
 use std::collections::HashSet;
 use std::io::Seek;
 use std::io::SeekFrom;
 use std::io::Write as IoWrite;
 use std::path::{Path, PathBuf};
 
 /// A repository on disk
 pub struct Repo {
     working_directory: PathBuf,
     dot_hg: PathBuf,
     store: PathBuf,
     requirements: HashSet<String>,
     config: Config,
     dirstate_parents: LazyCell<DirstateParents>,
     dirstate_data_file_uuid: LazyCell<Option<Vec<u8>>>,
     dirstate_map: LazyCell<OwningDirstateMap>,
     changelog: LazyCell<Changelog>,
     manifestlog: LazyCell<Manifestlog>,
 }
 
 #[derive(Debug, derive_more::From)]
 pub enum RepoError {
     NotFound {
         at: PathBuf,
     },
     #[from]
     ConfigParseError(ConfigParseError),
     #[from]
     Other(HgError),
 }
 
 impl From<ConfigError> for RepoError {
     fn from(error: ConfigError) -> Self {
         match error {
             ConfigError::Parse(error) => error.into(),
             ConfigError::Other(error) => error.into(),
         }
     }
 }
 
 impl Repo {
     /// tries to find nearest repository root in current working directory or
     /// its ancestors
     pub fn find_repo_root() -> Result<PathBuf, RepoError> {
         let current_directory = crate::utils::current_dir()?;
         // ancestors() is inclusive: it first yields `current_directory`
         // as-is.
         for ancestor in current_directory.ancestors() {
             if is_dir(ancestor.join(".hg"))? {
                 return Ok(ancestor.to_path_buf());
             }
         }
         return Err(RepoError::NotFound {
             at: current_directory,
         });
     }
 
     /// Find a repository, either at the given path (which must contain a `.hg`
     /// sub-directory) or by searching the current directory and its
     /// ancestors.
     ///
     /// A method with two very different "modes" like this usually a code smell
     /// to make two methods instead, but in this case an `Option` is what rhg
     /// sub-commands get from Clap for the `-R` / `--repository` CLI argument.
     /// Having two methods would just move that `if` to almost all callers.
     pub fn find(
         config: &Config,
         explicit_path: Option<PathBuf>,
     ) -> Result<Self, RepoError> {
         if let Some(root) = explicit_path {
             if is_dir(root.join(".hg"))? {
                 Self::new_at_path(root.to_owned(), config)
             } else if is_file(&root)? {
                 Err(HgError::unsupported("bundle repository").into())
             } else {
                 Err(RepoError::NotFound {
                     at: root.to_owned(),
                 })
             }
         } else {
             let root = Self::find_repo_root()?;
             Self::new_at_path(root, config)
         }
     }
 
     /// To be called after checking that `.hg` is a sub-directory
     fn new_at_path(
         working_directory: PathBuf,
         config: &Config,
     ) -> Result<Self, RepoError> {
         let dot_hg = working_directory.join(".hg");
 
         let mut repo_config_files = Vec::new();
         repo_config_files.push(dot_hg.join("hgrc"));
         repo_config_files.push(dot_hg.join("hgrc-not-shared"));
 
         let hg_vfs = Vfs { base: &dot_hg };
         let mut reqs = requirements::load_if_exists(hg_vfs)?;
         let relative =
             reqs.contains(requirements::RELATIVE_SHARED_REQUIREMENT);
         let shared =
             reqs.contains(requirements::SHARED_REQUIREMENT) || relative;
 
         // From `mercurial/localrepo.py`:
         //
         // if .hg/requires contains the sharesafe requirement, it means
         // there exists a `.hg/store/requires` too and we should read it
         // NOTE: presence of SHARESAFE_REQUIREMENT imply that store requirement
         // is present. We never write SHARESAFE_REQUIREMENT for a repo if store
         // is not present, refer checkrequirementscompat() for that
         //
         // However, if SHARESAFE_REQUIREMENT is not present, it means that the
         // repository was shared the old way. We check the share source
         // .hg/requires for SHARESAFE_REQUIREMENT to detect whether the
         // current repository needs to be reshared
         let share_safe = reqs.contains(requirements::SHARESAFE_REQUIREMENT);
 
         let store_path;
         if !shared {
             store_path = dot_hg.join("store");
         } else {
             let bytes = hg_vfs.read("sharedpath")?;
             let mut shared_path =
                 get_path_from_bytes(bytes.trim_end_matches(|b| b == b'\n'))
                     .to_owned();
             if relative {
                 shared_path = dot_hg.join(shared_path)
             }
             if !is_dir(&shared_path)? {
                 return Err(HgError::corrupted(format!(
                     ".hg/sharedpath points to nonexistent directory {}",
                     shared_path.display()
                 ))
                 .into());
             }
 
             store_path = shared_path.join("store");
 
             let source_is_share_safe =
                 requirements::load(Vfs { base: &shared_path })?
                     .contains(requirements::SHARESAFE_REQUIREMENT);
 
             if share_safe != source_is_share_safe {
                 return Err(HgError::unsupported("share-safe mismatch").into());
             }
 
             if share_safe {
                 repo_config_files.insert(0, shared_path.join("hgrc"))
             }
         }
         if share_safe {
             reqs.extend(requirements::load(Vfs { base: &store_path })?);
         }
 
         let repo_config = if std::env::var_os("HGRCSKIPREPO").is_none() {
             config.combine_with_repo(&repo_config_files)?
         } else {
             config.clone()
         };
 
         let repo = Self {
             requirements: reqs,
             working_directory,
             store: store_path,
             dot_hg,
             config: repo_config,
             dirstate_parents: LazyCell::new(),
             dirstate_data_file_uuid: LazyCell::new(),
             dirstate_map: LazyCell::new(),
             changelog: LazyCell::new(),
             manifestlog: LazyCell::new(),
         };
 
         requirements::check(&repo)?;
 
         Ok(repo)
     }
 
     pub fn working_directory_path(&self) -> &Path {
         &self.working_directory
     }
 
     pub fn requirements(&self) -> &HashSet<String> {
         &self.requirements
     }
 
     pub fn config(&self) -> &Config {
         &self.config
     }
 
     /// For accessing repository files (in `.hg`), except for the store
     /// (`.hg/store`).
     pub fn hg_vfs(&self) -> Vfs<'_> {
         Vfs { base: &self.dot_hg }
     }
 
     /// For accessing repository store files (in `.hg/store`)
     pub fn store_vfs(&self) -> Vfs<'_> {
         Vfs { base: &self.store }
     }
 
     /// For accessing the working copy
     pub fn working_directory_vfs(&self) -> Vfs<'_> {
         Vfs {
             base: &self.working_directory,
         }
     }
 
     pub fn try_with_wlock_no_wait<R>(
         &self,
         f: impl FnOnce() -> R,
     ) -> Result<R, LockError> {
         try_with_lock_no_wait(self.hg_vfs(), "wlock", f)
     }
 
     pub fn has_dirstate_v2(&self) -> bool {
         self.requirements
             .contains(requirements::DIRSTATE_V2_REQUIREMENT)
     }
 
     pub fn has_sparse(&self) -> bool {
         self.requirements.contains(requirements::SPARSE_REQUIREMENT)
     }
 
     pub fn has_narrow(&self) -> bool {
         self.requirements.contains(requirements::NARROW_REQUIREMENT)
     }
 
     pub fn has_nodemap(&self) -> bool {
         self.requirements
             .contains(requirements::NODEMAP_REQUIREMENT)
     }
 
     fn dirstate_file_contents(&self) -> Result<Vec<u8>, HgError> {
         Ok(self
             .hg_vfs()
             .read("dirstate")
             .io_not_found_as_none()?
             .unwrap_or(Vec::new()))
     }
 
     pub fn dirstate_parents(&self) -> Result<DirstateParents, HgError> {
         Ok(*self
             .dirstate_parents
             .get_or_init(|| self.read_dirstate_parents())?)
     }
 
     fn read_dirstate_parents(&self) -> Result<DirstateParents, HgError> {
         let dirstate = self.dirstate_file_contents()?;
         let parents = if dirstate.is_empty() {
             if self.has_dirstate_v2() {
                 self.dirstate_data_file_uuid.set(None);
             }
             DirstateParents::NULL
         } else if self.has_dirstate_v2() {
             let docket =
                 crate::dirstate_tree::on_disk::read_docket(&dirstate)?;
             self.dirstate_data_file_uuid
                 .set(Some(docket.uuid.to_owned()));
             docket.parents()
         } else {
             crate::dirstate::parsers::parse_dirstate_parents(&dirstate)?
                 .clone()
         };
         self.dirstate_parents.set(parents);
         Ok(parents)
     }
 
     fn read_dirstate_data_file_uuid(
         &self,
     ) -> Result<Option<Vec<u8>>, HgError> {
         assert!(
             self.has_dirstate_v2(),
             "accessing dirstate data file ID without dirstate-v2"
         );
         let dirstate = self.dirstate_file_contents()?;
         if dirstate.is_empty() {
             self.dirstate_parents.set(DirstateParents::NULL);
             Ok(None)
         } else {
             let docket =
                 crate::dirstate_tree::on_disk::read_docket(&dirstate)?;
             self.dirstate_parents.set(docket.parents());
             Ok(Some(docket.uuid.to_owned()))
         }
     }
 
     fn new_dirstate_map(&self) -> Result<OwningDirstateMap, DirstateError> {
         let dirstate_file_contents = self.dirstate_file_contents()?;
         if dirstate_file_contents.is_empty() {
             self.dirstate_parents.set(DirstateParents::NULL);
             if self.has_dirstate_v2() {
                 self.dirstate_data_file_uuid.set(None);
             }
             Ok(OwningDirstateMap::new_empty(Vec::new()))
         } else if self.has_dirstate_v2() {
             let docket = crate::dirstate_tree::on_disk::read_docket(
                 &dirstate_file_contents,
             )?;
             self.dirstate_parents.set(docket.parents());
             self.dirstate_data_file_uuid
                 .set(Some(docket.uuid.to_owned()));
             let data_size = docket.data_size();
             let metadata = docket.tree_metadata();
-            if crate::vfs::is_on_nfs_mount(docket.data_filename()) {
-                // Don't mmap on NFS to prevent `SIGBUS` error on deletion
-                OwningDirstateMap::new_v2(
-                    self.hg_vfs().read(docket.data_filename())?,
-                    data_size,
-                    metadata,
-                )
-            } else if let Some(data_mmap) = self
-                .hg_vfs()
-                .mmap_open(docket.data_filename())
-                .io_not_found_as_none()?
-            {
-                OwningDirstateMap::new_v2(data_mmap, data_size, metadata)
-            } else {
-                OwningDirstateMap::new_v2(Vec::new(), data_size, metadata)
-            }
+            let mut map =
+                if crate::vfs::is_on_nfs_mount(docket.data_filename()) {
+                    // Don't mmap on NFS to prevent `SIGBUS` error on deletion
+                    OwningDirstateMap::new_v2(
+                        self.hg_vfs().read(docket.data_filename())?,
+                        data_size,
+                        metadata,
+                    )
+                } else if let Some(data_mmap) = self
+                    .hg_vfs()
+                    .mmap_open(docket.data_filename())
+                    .io_not_found_as_none()?
+                {
+                    OwningDirstateMap::new_v2(data_mmap, data_size, metadata)
+                } else {
+                    OwningDirstateMap::new_v2(Vec::new(), data_size, metadata)
+                }?;
+
+            let write_mode_config = self
+                .config()
+                .get_str(b"devel", b"dirstate.v2.data_update_mode")
+                .unwrap_or(Some("auto"))
+                .unwrap_or("auto"); // don't bother for devel options
+            let write_mode = match write_mode_config {
+                "auto" => DirstateMapWriteMode::Auto,
+                "force-new" => DirstateMapWriteMode::ForceNewDataFile,
+                "force-append" => DirstateMapWriteMode::ForceAppend,
+                _ => DirstateMapWriteMode::Auto,
+            };
+
+            map.with_dmap_mut(|m| m.set_write_mode(write_mode));
+
+            Ok(map)
         } else {
             let (map, parents) =
                 OwningDirstateMap::new_v1(dirstate_file_contents)?;
             self.dirstate_parents.set(parents);
             Ok(map)
         }
     }
 
     pub fn dirstate_map(
         &self,
     ) -> Result<Ref<OwningDirstateMap>, DirstateError> {
         self.dirstate_map.get_or_init(|| self.new_dirstate_map())
     }
 
     pub fn dirstate_map_mut(
         &self,
     ) -> Result<RefMut<OwningDirstateMap>, DirstateError> {
         self.dirstate_map
             .get_mut_or_init(|| self.new_dirstate_map())
     }
 
     fn new_changelog(&self) -> Result<Changelog, HgError> {
         Changelog::open(&self.store_vfs(), self.has_nodemap())
     }
 
     pub fn changelog(&self) -> Result<Ref<Changelog>, HgError> {
         self.changelog.get_or_init(|| self.new_changelog())
     }
 
     pub fn changelog_mut(&self) -> Result<RefMut<Changelog>, HgError> {
         self.changelog.get_mut_or_init(|| self.new_changelog())
     }
 
     fn new_manifestlog(&self) -> Result<Manifestlog, HgError> {
         Manifestlog::open(&self.store_vfs(), self.has_nodemap())
     }
 
     pub fn manifestlog(&self) -> Result<Ref<Manifestlog>, HgError> {
         self.manifestlog.get_or_init(|| self.new_manifestlog())
     }
 
     pub fn manifestlog_mut(&self) -> Result<RefMut<Manifestlog>, HgError> {
         self.manifestlog.get_mut_or_init(|| self.new_manifestlog())
     }
 
     /// Returns the manifest of the *changeset* with the given node ID
     pub fn manifest_for_node(
         &self,
         node: impl Into<NodePrefix>,
     ) -> Result<Manifest, RevlogError> {
         self.manifestlog()?.data_for_node(
             self.changelog()?
                 .data_for_node(node.into())?
                 .manifest_node()?
                 .into(),
         )
     }
 
     /// Returns the manifest of the *changeset* with the given revision number
     pub fn manifest_for_rev(
         &self,
         revision: Revision,
     ) -> Result<Manifest, RevlogError> {
         self.manifestlog()?.data_for_node(
             self.changelog()?
                 .data_for_rev(revision)?
                 .manifest_node()?
                 .into(),
         )
     }
 
     pub fn has_subrepos(&self) -> Result<bool, DirstateError> {
         if let Some(entry) = self.dirstate_map()?.get(HgPath::new(".hgsub"))? {
             Ok(entry.tracked())
         } else {
             Ok(false)
         }
     }
 
     pub fn filelog(&self, path: &HgPath) -> Result<Filelog, HgError> {
         Filelog::open(self, path)
     }
 
     /// Write to disk any updates that were made through `dirstate_map_mut`.
     ///
     /// The "wlock" must be held while calling this.
     /// See for example `try_with_wlock_no_wait`.
     ///
     /// TODO: have a `WritableRepo` type only accessible while holding the
     /// lock?
     pub fn write_dirstate(&self) -> Result<(), DirstateError> {
         let map = self.dirstate_map()?;
         // TODO: Maintain a `DirstateMap::dirty` flag, and return early here if
         // it’s unset
         let parents = self.dirstate_parents()?;
         let (packed_dirstate, old_uuid_to_remove) = if self.has_dirstate_v2() {
             let uuid_opt = self
                 .dirstate_data_file_uuid
                 .get_or_init(|| self.read_dirstate_data_file_uuid())?;
             let uuid_opt = uuid_opt.as_ref();
             let write_mode = if uuid_opt.is_some() {
                 DirstateMapWriteMode::Auto
             } else {
                 DirstateMapWriteMode::ForceNewDataFile
             };
             let (data, tree_metadata, append, old_data_size) =
                 map.pack_v2(write_mode)?;
 
             // Reuse the uuid, or generate a new one, keeping the old for
             // deletion.
             let (uuid, old_uuid) = match uuid_opt {
                 Some(uuid) => {
                     let as_str = std::str::from_utf8(uuid)
                         .map_err(|_| {
                             HgError::corrupted(
                                 "non-UTF-8 dirstate data file ID",
                             )
                         })?
                         .to_owned();
                     if append {
                         (as_str, None)
                     } else {
                         (DirstateDocket::new_uid(), Some(as_str))
                     }
                 }
                 None => (DirstateDocket::new_uid(), None),
             };
 
             let data_filename = format!("dirstate.{}", uuid);
             let data_filename = self.hg_vfs().join(data_filename);
             let mut options = std::fs::OpenOptions::new();
             options.write(true);
 
             // Why are we not using the O_APPEND flag when appending?
             //
             // - O_APPEND makes it trickier to deal with garbage at the end of
             //   the file, left by a previous uncommitted transaction. By
             //   starting the write at [old_data_size] we make sure we erase
             //   all such garbage.
             //
             // - O_APPEND requires to special-case 0-byte writes, whereas we
             //   don't need that.
             //
             // - Some OSes have bugs in implementation O_APPEND:
             //   revlog.py talks about a Solaris bug, but we also saw some ZFS
             //   bug: https://github.com/openzfs/zfs/pull/3124,
             //   https://github.com/openzfs/zfs/issues/13370
             //
             if !append {
                 log::trace!("creating a new dirstate data file");
                 options.create_new(true);
             } else {
                 log::trace!("appending to the dirstate data file");
             }
 
             let data_size = (|| {
                 // TODO: loop and try another random ID if !append and this
                 // returns `ErrorKind::AlreadyExists`? Collision chance of two
                 // random IDs is one in 2**32
                 let mut file = options.open(&data_filename)?;
                 if append {
                     file.seek(SeekFrom::Start(old_data_size as u64))?;
                 }
                 file.write_all(&data)?;
                 file.flush()?;
                 file.seek(SeekFrom::Current(0))
             })()
             .when_writing_file(&data_filename)?;
 
             let packed_dirstate = DirstateDocket::serialize(
                 parents,
                 tree_metadata,
                 data_size,
                 uuid.as_bytes(),
             )
             .map_err(|_: std::num::TryFromIntError| {
                 HgError::corrupted("overflow in dirstate docket serialization")
             })?;
 
             (packed_dirstate, old_uuid)
         } else {
             (map.pack_v1(parents)?, None)
         };
 
         let vfs = self.hg_vfs();
         vfs.atomic_write("dirstate", &packed_dirstate)?;
         if let Some(uuid) = old_uuid_to_remove {
             // Remove the old data file after the new docket pointing to the
             // new data file was written.
             vfs.remove_file(format!("dirstate.{}", uuid))?;
         }
         Ok(())
     }
 }
 
 /// Lazily-initialized component of `Repo` with interior mutability
 ///
 /// This differs from `OnceCell` in that the value can still be "deinitialized"
 /// later by setting its inner `Option` to `None`. It also takes the
 /// initialization function as an argument when the value is requested, not
 /// when the instance is created.
 struct LazyCell<T> {
     value: RefCell<Option<T>>,
 }
 
 impl<T> LazyCell<T> {
     fn new() -> Self {
         Self {
             value: RefCell::new(None),
         }
     }
 
     fn set(&self, value: T) {
         *self.value.borrow_mut() = Some(value)
     }
 
     fn get_or_init<E>(
         &self,
         init: impl Fn() -> Result<T, E>,
     ) -> Result<Ref<T>, E> {
         let mut borrowed = self.value.borrow();
         if borrowed.is_none() {
             drop(borrowed);
             // Only use `borrow_mut` if it is really needed to avoid panic in
             // case there is another outstanding borrow but mutation is not
             // needed.
             *self.value.borrow_mut() = Some(init()?);
             borrowed = self.value.borrow()
         }
         Ok(Ref::map(borrowed, |option| option.as_ref().unwrap()))
     }
 
     fn get_mut_or_init<E>(
         &self,
         init: impl Fn() -> Result<T, E>,
     ) -> Result<RefMut<T>, E> {
         let mut borrowed = self.value.borrow_mut();
         if borrowed.is_none() {
             *borrowed = Some(init()?);
         }
         Ok(RefMut::map(borrowed, |option| option.as_mut().unwrap()))
     }
 }

rust/hg-cpython/src/dirstate/dirstate_map.rs

@@ -1,550 +1,551 b''
 // dirstate_map.rs
 //
 // Copyright 2019 Raphaël Gomès <[e-mail]>
 //
 // This software may be used and distributed according to the terms of the
 // GNU General Public License version 2 or any later version.
 
 //! Bindings for the `hg::dirstate::dirstate_map` file provided by the
 //! `hg-core` package.
 
 use std::cell::{RefCell, RefMut};
 use std::convert::TryInto;
 
 use cpython::{
     exc, PyBool, PyBytes, PyClone, PyDict, PyErr, PyList, PyNone, PyObject,
     PyResult, Python, PythonObject, ToPyObject, UnsafePyLeaked,
 };
 use hg::dirstate::{ParentFileData, TruncatedTimestamp};
 
 use crate::{
     dirstate::copymap::{CopyMap, CopyMapItemsIterator, CopyMapKeysIterator},
     dirstate::item::DirstateItem,
     pybytes_deref::PyBytesDeref,
 };
 use hg::{
     dirstate::StateMapIter, dirstate_tree::dirstate_map::DirstateMapWriteMode,
     dirstate_tree::on_disk::DirstateV2ParseError,
     dirstate_tree::owning::OwningDirstateMap, revlog::Node,
     utils::files::normalize_case, utils::hg_path::HgPath, DirstateEntry,
     DirstateError, DirstateParents,
 };
 
 // TODO
 //     This object needs to share references to multiple members of its Rust
 //     inner struct, namely `copy_map`, `dirs` and `all_dirs`.
 //     Right now `CopyMap` is done, but it needs to have an explicit reference
 //     to `RustDirstateMap` which itself needs to have an encapsulation for
 //     every method in `CopyMap` (copymapcopy, etc.).
 //     This is ugly and hard to maintain.
 //     The same logic applies to `dirs` and `all_dirs`, however the `Dirs`
 //     `py_class!` is already implemented and does not mention
 //     `RustDirstateMap`, rightfully so.
 //     All attributes also have to have a separate refcount data attribute for
 //     leaks, with all methods that go along for reference sharing.
 py_class!(pub class DirstateMap |py| {
     @shared data inner: OwningDirstateMap;
 
     /// Returns a `(dirstate_map, parents)` tuple
     @staticmethod
     def new_v1(
         on_disk: PyBytes,
     ) -> PyResult<PyObject> {
         let on_disk = PyBytesDeref::new(py, on_disk);
         let (map, parents) = OwningDirstateMap::new_v1(on_disk)
             .map_err(|e| dirstate_error(py, e))?;
         let map = Self::create_instance(py, map)?;
         let p1 = PyBytes::new(py, parents.p1.as_bytes());
         let p2 = PyBytes::new(py, parents.p2.as_bytes());
         let parents = (p1, p2);
         Ok((map, parents).to_py_object(py).into_object())
     }
 
     /// Returns a DirstateMap
     @staticmethod
     def new_v2(
         on_disk: PyBytes,
         data_size: usize,
         tree_metadata: PyBytes,
     ) -> PyResult<PyObject> {
         let dirstate_error = |e: DirstateError| {
             PyErr::new::<exc::OSError, _>(py, format!("Dirstate error: {:?}", e))
         };
         let on_disk = PyBytesDeref::new(py, on_disk);
         let map = OwningDirstateMap::new_v2(
             on_disk, data_size, tree_metadata.data(py),
         ).map_err(dirstate_error)?;
         let map = Self::create_instance(py, map)?;
         Ok(map.into_object())
     }
 
     def clear(&self) -> PyResult<PyObject> {
         self.inner(py).borrow_mut().clear();
         Ok(py.None())
     }
 
     def get(
         &self,
         key: PyObject,
         default: Option<PyObject> = None
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(entry) => {
                 Ok(Some(DirstateItem::new_as_pyobject(py, entry)?))
             },
             None => Ok(default)
         }
     }
 
     def set_tracked(&self, f: PyObject) -> PyResult<PyBool> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_tracked(path);
         let was_tracked = res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(was_tracked.to_py_object(py))
     }
 
     def set_untracked(&self, f: PyObject) -> PyResult<PyBool> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_untracked(path);
         let was_tracked = res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(was_tracked.to_py_object(py))
     }
 
     def set_clean(
         &self,
         f: PyObject,
         mode: u32,
         size: u32,
         mtime: (i64, u32, bool)
     ) -> PyResult<PyNone> {
         let (mtime_s, mtime_ns, second_ambiguous) = mtime;
         let timestamp = TruncatedTimestamp::new_truncate(
             mtime_s, mtime_ns, second_ambiguous
         );
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_clean(
             path, mode, size, timestamp,
         );
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def set_possibly_dirty(&self, f: PyObject) -> PyResult<PyNone> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_possibly_dirty(path);
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def reset_state(
         &self,
         f: PyObject,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parentfiledata: Option<(u32, u32, Option<(i64, u32, bool)>)>,
     ) -> PyResult<PyNone> {
         let mut has_meaningful_mtime = has_meaningful_mtime;
         let parent_file_data = match parentfiledata {
             None => {
                 has_meaningful_mtime = false;
                 None
             },
             Some(data) => {
                 let (mode, size, mtime_info) = data;
                 let mtime = if let Some(mtime_info) = mtime_info {
                     let (mtime_s, mtime_ns, second_ambiguous) = mtime_info;
                     let timestamp = TruncatedTimestamp::new_truncate(
                         mtime_s, mtime_ns, second_ambiguous
                     );
                     Some(timestamp)
                 } else {
                     has_meaningful_mtime = false;
                     None
                 };
                 Some(ParentFileData {
                     mode_size: Some((mode, size)),
                     mtime,
                 })
             }
         };
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().reset_state(
             path,
             wc_tracked,
             p1_tracked,
             p2_info,
             has_meaningful_mtime,
             parent_file_data,
         );
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def hastrackeddir(&self, d: PyObject) -> PyResult<PyBool> {
         let d = d.extract::<PyBytes>(py)?;
         Ok(self.inner(py).borrow_mut()
             .has_tracked_dir(HgPath::new(d.data(py)))
             .map_err(|e| {
                 PyErr::new::<exc::ValueError, _>(py, e.to_string())
             })?
             .to_py_object(py))
     }
 
     def hasdir(&self, d: PyObject) -> PyResult<PyBool> {
         let d = d.extract::<PyBytes>(py)?;
         Ok(self.inner(py).borrow_mut()
             .has_dir(HgPath::new(d.data(py)))
             .map_err(|e| {
                 PyErr::new::<exc::ValueError, _>(py, e.to_string())
             })?
             .to_py_object(py))
     }
 
     def write_v1(
         &self,
         p1: PyObject,
         p2: PyObject,
     ) -> PyResult<PyBytes> {
         let inner = self.inner(py).borrow();
         let parents = DirstateParents {
             p1: extract_node_id(py, &p1)?,
             p2: extract_node_id(py, &p2)?,
         };
         let result = inner.pack_v1(parents);
         match result {
             Ok(packed) => Ok(PyBytes::new(py, &packed)),
             Err(_) => Err(PyErr::new::<exc::OSError, _>(
                 py,
                 "Dirstate error".to_string(),
             )),
         }
     }
 
     /// Returns new data together with whether that data should be appended to
     /// the existing data file whose content is at `self.on_disk` (True),
     /// instead of written to a new data file (False).
     def write_v2(
         &self,
         write_mode: usize,
     ) -> PyResult<PyObject> {
         let inner = self.inner(py).borrow();
         let rust_write_mode = match write_mode {
             0 => DirstateMapWriteMode::Auto,
             1 => DirstateMapWriteMode::ForceNewDataFile,
+            2 => DirstateMapWriteMode::ForceAppend,
             _ => DirstateMapWriteMode::Auto, // XXX should we error out?
         };
         let result = inner.pack_v2(rust_write_mode);
         match result {
             Ok((packed, tree_metadata, append, _old_data_size)) => {
                 let packed = PyBytes::new(py, &packed);
                 let tree_metadata = PyBytes::new(py, tree_metadata.as_bytes());
                 let tuple = (packed, tree_metadata, append);
                 Ok(tuple.to_py_object(py).into_object())
             },
             Err(_) => Err(PyErr::new::<exc::OSError, _>(
                 py,
                 "Dirstate error".to_string(),
             )),
         }
     }
 
     def filefoldmapasdict(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         for item in self.inner(py).borrow_mut().iter() {
             let (path, entry) = item.map_err(|e| v2_error(py, e))?;
             if !entry.removed() {
                 let key = normalize_case(path);
                 let value = path;
                 dict.set_item(
                     py,
                     PyBytes::new(py, key.as_bytes()).into_object(),
                     PyBytes::new(py, value.as_bytes()).into_object(),
                 )?;
             }
         }
         Ok(dict)
     }
 
     def __len__(&self) -> PyResult<usize> {
         Ok(self.inner(py).borrow().len())
     }
 
     def __contains__(&self, key: PyObject) -> PyResult<bool> {
         let key = key.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow()
             .contains_key(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))
     }
 
     def __getitem__(&self, key: PyObject) -> PyResult<PyObject> {
         let key = key.extract::<PyBytes>(py)?;
         let key = HgPath::new(key.data(py));
         match self
             .inner(py)
             .borrow()
             .get(key)
             .map_err(|e| v2_error(py, e))?
         {
             Some(entry) => {
                 Ok(DirstateItem::new_as_pyobject(py, entry)?)
             },
             None => Err(PyErr::new::<exc::KeyError, _>(
                 py,
                 String::from_utf8_lossy(key.as_bytes()),
             )),
         }
     }
 
     def keys(&self) -> PyResult<DirstateMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     def items(&self) -> PyResult<DirstateMapItemsIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapItemsIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     def __iter__(&self) -> PyResult<DirstateMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     // TODO all copymap* methods, see docstring above
     def copymapcopy(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         for item in self.inner(py).borrow().copy_map_iter() {
             let (key, value) = item.map_err(|e| v2_error(py, e))?;
             dict.set_item(
                 py,
                 PyBytes::new(py, key.as_bytes()),
                 PyBytes::new(py, value.as_bytes()),
             )?;
         }
         Ok(dict)
     }
 
     def copymapgetitem(&self, key: PyObject) -> PyResult<PyBytes> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .copy_map_get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(PyBytes::new(py, copy.as_bytes())),
             None => Err(PyErr::new::<exc::KeyError, _>(
                 py,
                 String::from_utf8_lossy(key.data(py)),
             )),
         }
     }
     def copymap(&self) -> PyResult<CopyMap> {
         CopyMap::from_inner(py, self.clone_ref(py))
     }
 
     def copymaplen(&self) -> PyResult<usize> {
         Ok(self.inner(py).borrow().copy_map_len())
     }
     def copymapcontains(&self, key: PyObject) -> PyResult<bool> {
         let key = key.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow()
             .copy_map_contains_key(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))
     }
     def copymapget(
         &self,
         key: PyObject,
         default: Option<PyObject>
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .copy_map_get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(Some(
                 PyBytes::new(py, copy.as_bytes()).into_object(),
             )),
             None => Ok(default),
         }
     }
     def copymapsetitem(
         &self,
         key: PyObject,
         value: PyObject
     ) -> PyResult<PyObject> {
         let key = key.extract::<PyBytes>(py)?;
         let value = value.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow_mut()
             .copy_map_insert(
                 HgPath::new(key.data(py)),
                 HgPath::new(value.data(py)),
             )
             .map_err(|e| v2_error(py, e))?;
         Ok(py.None())
     }
     def copymappop(
         &self,
         key: PyObject,
         default: Option<PyObject>
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow_mut()
             .copy_map_remove(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(Some(
                 PyBytes::new(py, copy.as_bytes()).into_object(),
             )),
             None => Ok(default),
         }
     }
 
     def copymapiter(&self) -> PyResult<CopyMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         CopyMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.copy_map_iter()) },
         )
     }
 
     def copymapitemsiter(&self) -> PyResult<CopyMapItemsIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         CopyMapItemsIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.copy_map_iter()) },
         )
     }
 
     def tracked_dirs(&self) -> PyResult<PyList> {
         let dirs = PyList::new(py, &[]);
         for path in self.inner(py).borrow_mut().iter_tracked_dirs()
             .map_err(|e |dirstate_error(py, e))?
         {
             let path = path.map_err(|e| v2_error(py, e))?;
             let path = PyBytes::new(py, path.as_bytes());
             dirs.append(py, path.into_object())
         }
         Ok(dirs)
     }
 
     def setparents_fixup(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         let copies = self.inner(py).borrow_mut().setparents_fixup();
         for (key, value) in copies.map_err(|e| v2_error(py, e))? {
             dict.set_item(
                 py,
                 PyBytes::new(py, key.as_bytes()),
                 PyBytes::new(py, value.as_bytes()),
             )?;
         }
         Ok(dict)
     }
 
     def debug_iter(&self, all: bool) -> PyResult<PyList> {
         let dirs = PyList::new(py, &[]);
         for item in self.inner(py).borrow().debug_iter(all) {
             let (path, (state, mode, size, mtime)) =
                 item.map_err(|e| v2_error(py, e))?;
             let path = PyBytes::new(py, path.as_bytes());
             let item = (path, state, mode, size, mtime);
             dirs.append(py, item.to_py_object(py).into_object())
         }
         Ok(dirs)
     }
 });
 
 impl DirstateMap {
     pub fn get_inner_mut<'a>(
         &'a self,
         py: Python<'a>,
     ) -> RefMut<'a, OwningDirstateMap> {
         self.inner(py).borrow_mut()
     }
     fn translate_key(
         py: Python,
         res: Result<(&HgPath, DirstateEntry), DirstateV2ParseError>,
     ) -> PyResult<Option<PyBytes>> {
         let (f, _entry) = res.map_err(|e| v2_error(py, e))?;
         Ok(Some(PyBytes::new(py, f.as_bytes())))
     }
     fn translate_key_value(
         py: Python,
         res: Result<(&HgPath, DirstateEntry), DirstateV2ParseError>,
     ) -> PyResult<Option<(PyBytes, PyObject)>> {
         let (f, entry) = res.map_err(|e| v2_error(py, e))?;
         Ok(Some((
             PyBytes::new(py, f.as_bytes()),
             DirstateItem::new_as_pyobject(py, entry)?,
         )))
     }
 }
 
 py_shared_iterator!(
     DirstateMapKeysIterator,
     UnsafePyLeaked<StateMapIter<'static>>,
     DirstateMap::translate_key,
     Option<PyBytes>
 );
 
 py_shared_iterator!(
     DirstateMapItemsIterator,
     UnsafePyLeaked<StateMapIter<'static>>,
     DirstateMap::translate_key_value,
     Option<(PyBytes, PyObject)>
 );
 
 fn extract_node_id(py: Python, obj: &PyObject) -> PyResult<Node> {
     let bytes = obj.extract::<PyBytes>(py)?;
     match bytes.data(py).try_into() {
         Ok(s) => Ok(s),
         Err(e) => Err(PyErr::new::<exc::ValueError, _>(py, e.to_string())),
     }
 }
 
 pub(super) fn v2_error(py: Python<'_>, _: DirstateV2ParseError) -> PyErr {
     PyErr::new::<exc::ValueError, _>(py, "corrupted dirstate-v2")
 }
 
 fn dirstate_error(py: Python<'_>, e: DirstateError) -> PyErr {
     PyErr::new::<exc::OSError, _>(py, format!("Dirstate error: {:?}", e))
 }

tests/test-dirstate-status-write-race.t

@@ -1,433 +1,433 b''
 =====================================================================
 Check potential race conditions between a status and other operations
 =====================================================================
 
 #testcases dirstate-v1 dirstate-v2
 
 The `hg status` command can run without the wlock, however it might end up
 having to update the on-disk dirstate files, for example to mark ambiguous
 files as clean, or to update directory caches information with dirstate-v2.
 
 
 If another process updates the dirstate in the meantime we might run into
 trouble. Especially, commands doing semantic changes like `hg add` or
  `hg commit` should not see their update erased by a concurrent status.
 
 Unlike commands like `add` or `commit`, `status` only writes the dirstate
 to update caches, no actual information is lost if we fail to write to disk.
 
 
 This test file is meant to test various cases where such parallel operations
 between a status with reasons to update the dirstate and another semantic
 changes happen.
 
 
 Setup
 =====
 
   $ cat >> $HGRCPATH << EOF
   > [storage]
   > dirstate-v2.slow-path=allow
   > EOF
 
 #if dirstate-v2
   $ cat >> $HGRCPATH << EOF
   > [format]
   > use-dirstate-v2=yes
   > EOF
 #else
   $ cat >> $HGRCPATH << EOF
   > [format]
   > use-dirstate-v2=no
   > EOF
 #endif
 
   $ directories="dir dir/nested dir2"
   $ first_files="dir/nested/a dir/b dir/c dir/d dir2/e f"
   $ second_files="g dir/nested/h dir/i dir/j dir2/k dir2/l dir/nested/m"
   $ extra_files="dir/n dir/o p q"
 
   $ hg init reference-repo
   $ cd reference-repo
   $ mkdir -p dir/nested dir2
   $ touch -t 200001010000 $first_files $directories
   $ hg commit -Aqm "recreate a bunch of files to facilitate dirstate-v2 append"
   $ touch -t 200001010010 $second_files $directories
   $ hg commit -Aqm "more files to have two commits"
   $ hg log -G -v
   @  changeset:   1:c349430a1631
   |  tag:         tip
   |  user:        test
   |  date:        Thu Jan 01 00:00:00 1970 +0000
   |  files:       dir/i dir/j dir/nested/h dir/nested/m dir2/k dir2/l g
   |  description:
   |  more files to have two commits
   |
   |
   o  changeset:   0:4f23db756b09
      user:        test
      date:        Thu Jan 01 00:00:00 1970 +0000
      files:       dir/b dir/c dir/d dir/nested/a dir2/e f
      description:
      recreate a bunch of files to facilitate dirstate-v2 append
   
   
   $ hg manifest
   dir/b
   dir/c
   dir/d
   dir/i
   dir/j
   dir/nested/a
   dir/nested/h
   dir/nested/m
   dir2/e
   dir2/k
   dir2/l
   f
   g
 
 Add some unknown files and refresh the dirstate
 
   $ touch -t 200001010020 $extra_files
   $ hg add dir/o
   $ hg remove dir/nested/m
 
-  $ hg st
+  $ hg st --config devel.dirstate.v2.data_update_mode=force-new
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ hg debugstate
   n 644          0 2000-01-01 00:00:00 dir/b
   n 644          0 2000-01-01 00:00:00 dir/c
   n 644          0 2000-01-01 00:00:00 dir/d
   n 644          0 2000-01-01 00:10:00 dir/i
   n 644          0 2000-01-01 00:10:00 dir/j
   n 644          0 2000-01-01 00:00:00 dir/nested/a
   n 644          0 2000-01-01 00:10:00 dir/nested/h
   r ?????????????????????????????????? dir/nested/m (glob)
   a ?????????????????????????????????? dir/o (glob)
   n 644          0 2000-01-01 00:00:00 dir2/e
   n 644          0 2000-01-01 00:10:00 dir2/k
   n 644          0 2000-01-01 00:10:00 dir2/l
   n 644          0 2000-01-01 00:00:00 f
   n 644          0 2000-01-01 00:10:00 g
   $ hg debugstate > ../reference
   $ cd ..
 
 Explain / verify the test principles
 ------------------------------------
 
 First, we can properly copy the reference
 
   $ cp -a reference-repo sanity-check
   $ cd sanity-check
   $ hg debugstate
   n 644          0 2000-01-01 00:00:00 dir/b
   n 644          0 2000-01-01 00:00:00 dir/c
   n 644          0 2000-01-01 00:00:00 dir/d
   n 644          0 2000-01-01 00:10:00 dir/i
   n 644          0 2000-01-01 00:10:00 dir/j
   n 644          0 2000-01-01 00:00:00 dir/nested/a
   n 644          0 2000-01-01 00:10:00 dir/nested/h
   r ?????????????????????????????????? dir/nested/m (glob)
   a ?????????????????????????????????? dir/o (glob)
   n 644          0 2000-01-01 00:00:00 dir2/e
   n 644          0 2000-01-01 00:10:00 dir2/k
   n 644          0 2000-01-01 00:10:00 dir2/l
   n 644          0 2000-01-01 00:00:00 f
   n 644          0 2000-01-01 00:10:00 g
   $ hg debugstate > ../post-copy
   $ diff ../reference ../post-copy
 
 And status thinks the cache is in a proper state
 
   $ hg st
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ hg debugstate
   n 644          0 2000-01-01 00:00:00 dir/b
   n 644          0 2000-01-01 00:00:00 dir/c
   n 644          0 2000-01-01 00:00:00 dir/d
   n 644          0 2000-01-01 00:10:00 dir/i
   n 644          0 2000-01-01 00:10:00 dir/j
   n 644          0 2000-01-01 00:00:00 dir/nested/a
   n 644          0 2000-01-01 00:10:00 dir/nested/h
   r ?????????????????????????????????? dir/nested/m (glob)
   a ?????????????????????????????????? dir/o (glob)
   n 644          0 2000-01-01 00:00:00 dir2/e
   n 644          0 2000-01-01 00:10:00 dir2/k
   n 644          0 2000-01-01 00:10:00 dir2/l
   n 644          0 2000-01-01 00:00:00 f
   n 644          0 2000-01-01 00:10:00 g
   $ hg debugstate > ../post-status
   $ diff ../reference ../post-status
 
 Then we can start a status that:
 - has some update to do (the touch call)
 - will wait AFTER running status, but before updating the cache on disk
 
   $ touch -t 200001010001 dir/c
   $ hg st >$TESTTMP/status-race-lock.out 2>$TESTTMP/status-race-lock.log \
   > --config rhg.on-unsupported=abort \
   > --config devel.sync.status.pre-dirstate-write-file=$TESTTMP/status-race-lock \
   > &
   $ $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/status-race-lock.waiting
 
 We check it runs the status first by modifying a file and updating another timestamp
 
   $ touch -t 200001010003 dir/i
   $ echo babar > dir/j
   $ touch $TESTTMP/status-race-lock
   $ wait
 
 The test process should have reported a status before the change we made,
 and should have missed the timestamp update
 
   $ cat $TESTTMP/status-race-lock.out
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ cat $TESTTMP/status-race-lock.log
   $ hg debugstate | grep dir/c
   n 644          0 2000-01-01 00:01:00 dir/c
   $ hg debugstate | grep dir/i
   n 644          0 2000-01-01 00:10:00 dir/i
   $ hg debugstate | grep dir/j
   n 644          0 2000-01-01 00:10:00 dir/j
 
 final cleanup
 
   $ rm $TESTTMP/status-race-lock $TESTTMP/status-race-lock.waiting
   $ cd ..
 
 Actual Testing
 ==============
 
 Race with a `hg add`
 -------------------
 
   $ cp -a reference-repo race-with-add
   $ cd race-with-add
 
 spin a `hg status` with some caches to update
 
   $ touch -t 200001020001 f
   $ hg st >$TESTTMP/status-race-lock.out 2>$TESTTMP/status-race-lock.log \
   > --config rhg.on-unsupported=abort \
   > --config devel.sync.status.pre-dirstate-write-file=$TESTTMP/status-race-lock \
   > &
   $ $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/status-race-lock.waiting
 
 Add a file
 
   $ hg add dir/n
   $ touch $TESTTMP/status-race-lock
   $ wait
 
 The file should in a "added" state
 
   $ hg status
   A dir/n (no-rhg !)
   A dir/n (rhg no-dirstate-v1 !)
   A dir/n (missing-correct-output rhg dirstate-v1 !)
   A dir/o
   R dir/nested/m
   ? dir/n (known-bad-output rhg dirstate-v1 !)
   ? p
   ? q
 
 The status process should return a consistent result and not crash.
 
   $ cat $TESTTMP/status-race-lock.out
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ cat $TESTTMP/status-race-lock.log
   abort: when writing $TESTTMP/race-with-add/.hg/dirstate.*: $ENOENT$ (glob) (known-bad-output rhg dirstate-v2 !)
 
 final cleanup
 
   $ rm $TESTTMP/status-race-lock $TESTTMP/status-race-lock.waiting
   $ cd ..
 
 Race with a `hg commit`
 ----------------------
 
   $ cp -a reference-repo race-with-commit
   $ cd race-with-commit
 
 spin a `hg status` with some caches to update
 
   $ touch -t 200001020001 dir/j
   $ hg st >$TESTTMP/status-race-lock.out 2>$TESTTMP/status-race-lock.log \
   > --config rhg.on-unsupported=abort \
   > --config devel.sync.status.pre-dirstate-write-file=$TESTTMP/status-race-lock \
   > &
   $ $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/status-race-lock.waiting
 
 Add a file and force the data file rewrite
 
   $ hg commit -m created-during-status dir/o
   $ touch $TESTTMP/status-race-lock
   $ wait
 
 The parent must change and the status should be clean
 
 # XXX rhg misbehaves here
 #if no-rhg
   $ hg summary
   parent: 2:2e3b442a2fd4 tip
    created-during-status
   branch: default
   commit: 1 removed, 3 unknown
   update: (current)
   phases: 3 draft
   $ hg status
   R dir/nested/m
   ? dir/n
   ? p
   ? q
 #else
   $ hg summary
   parent: 1:c349430a1631 
    more files to have two commits
   branch: default
   commit: 1 added, 1 removed, 3 unknown (new branch head)
   update: 1 new changesets (update)
   phases: 3 draft
   $ hg status
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
 #endif
 
 The status process should return a consistent result and not crash.
 
   $ cat $TESTTMP/status-race-lock.out
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ cat $TESTTMP/status-race-lock.log
   abort: when removing $TESTTMP/race-with-commit/.hg/dirstate.*: $ENOENT$ (glob) (known-bad-output rhg dirstate-v2 !)
 
 final cleanup
 
   $ rm $TESTTMP/status-race-lock $TESTTMP/status-race-lock.waiting
   $ cd ..
 
 Race with a `hg update`
 ----------------------
 
   $ cp -a reference-repo race-with-update
   $ cd race-with-update
 
 spin a `hg status` with some caches to update
 
   $ touch -t 200001020001 dir2/k
   $ hg st >$TESTTMP/status-race-lock.out 2>$TESTTMP/status-race-lock.log \
   > --config rhg.on-unsupported=abort \
   > --config devel.sync.status.pre-dirstate-write-file=$TESTTMP/status-race-lock \
   > &
   $ $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/status-race-lock.waiting
 
 Add a file and force the data file rewrite
 
   $ hg update ".~1"
   0 files updated, 0 files merged, 6 files removed, 0 files unresolved
   $ touch $TESTTMP/status-race-lock
   $ wait
 
 The parent must change and the status should be clean
 
   $ hg summary
   parent: 0:4f23db756b09 
    recreate a bunch of files to facilitate dirstate-v2 append
   branch: default
   commit: 1 added, 3 unknown (new branch head)
   update: 1 new changesets (update)
   phases: 2 draft
   $ hg status
   A dir/o
   ? dir/n
   ? p
   ? q
 
 The status process should return a consistent result and not crash.
 
   $ cat $TESTTMP/status-race-lock.out
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ cat $TESTTMP/status-race-lock.log
   abort: when reading $TESTTMP/race-with-update/dir2/k: $ENOENT$ (known-bad-output rhg !)
 
 final cleanup
 
   $ rm $TESTTMP/status-race-lock $TESTTMP/status-race-lock.waiting
   $ cd ..
 
 Race with another status
 ------------------------
 
   $ cp -a reference-repo race-with-status
   $ cd race-with-status
 
 spin a `hg status` with some caches to update
 
   $ touch -t 200001010030 dir/nested/h
   $ hg st >$TESTTMP/status-race-lock.out 2>$TESTTMP/status-race-lock.log \
   > --config rhg.on-unsupported=abort \
   > --config devel.sync.status.pre-dirstate-write-file=$TESTTMP/status-race-lock \
   > &
   $ $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/status-race-lock.waiting
 
 touch g
 
   $ touch -t 200001010025 g
   $ hg status
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ touch $TESTTMP/status-race-lock
   $ wait
 
 the first update should be on disk
 
   $ hg debugstate --all | grep "g"
   n 644          0 2000-01-01 00:25:00 g (no-rhg !)
   n 644          0 2000-01-01 00:25:00 g (missing-correct-output rhg !)
   n 644          0 2000-01-01 00:10:00 g (known-bad-output rhg !)
 
 The status process should return a consistent result and not crash.
 
   $ cat $TESTTMP/status-race-lock.out
   A dir/o
   R dir/nested/m
   ? dir/n
   ? p
   ? q
   $ cat $TESTTMP/status-race-lock.log
   abort: when removing $TESTTMP/race-with-status/.hg/dirstate.*: $ENOENT$ (glob) (known-bad-output rhg dirstate-v2 !)
 
 final cleanup
 
   $ rm $TESTTMP/status-race-lock $TESTTMP/status-race-lock.waiting
   $ cd ..

tests/test-dirstate.t

@@ -1,259 +1,474 b''
 #testcases dirstate-v1 dirstate-v2
 
 #if dirstate-v2
   $ cat >> $HGRCPATH << EOF
   > [format]
   > use-dirstate-v2=1
   > [storage]
   > dirstate-v2.slow-path=allow
   > EOF
 #endif
 
 ------ Test dirstate._dirs refcounting
 
   $ hg init t
   $ cd t
   $ mkdir -p a/b/c/d
   $ touch a/b/c/d/x
   $ touch a/b/c/d/y
   $ touch a/b/c/d/z
   $ hg ci -Am m
   adding a/b/c/d/x
   adding a/b/c/d/y
   adding a/b/c/d/z
   $ hg mv a z
   moving a/b/c/d/x to z/b/c/d/x
   moving a/b/c/d/y to z/b/c/d/y
   moving a/b/c/d/z to z/b/c/d/z
 
 Test name collisions
 
   $ rm z/b/c/d/x
   $ mkdir z/b/c/d/x
   $ touch z/b/c/d/x/y
   $ hg add z/b/c/d/x/y
   abort: file 'z/b/c/d/x' in dirstate clashes with 'z/b/c/d/x/y'
   [255]
   $ rm -rf z/b/c/d
   $ touch z/b/c/d
   $ hg add z/b/c/d
   abort: directory 'z/b/c/d' already in dirstate
   [255]
 
   $ cd ..
 
 Issue1790: dirstate entry locked into unset if file mtime is set into
 the future
 
 Prepare test repo:
 
   $ hg init u
   $ cd u
   $ echo a > a
   $ hg add
   adding a
   $ hg ci -m1
 
 Set mtime of a into the future:
 
   $ touch -t 203101011200 a
 
 Status must not set a's entry to unset (issue1790):
 
   $ hg status
   $ hg debugstate
   n 644          2 2031-01-01 12:00:00 a
 
 Test modulo storage/comparison of absurd dates:
 
 #if no-aix
   $ touch -t 195001011200 a
   $ hg st
   $ hg debugstate
   n 644          2 2018-01-19 15:14:08 a
 #endif
 
 Verify that exceptions during a dirstate change leave the dirstate
 coherent (issue4353)
 
   $ cat > ../dirstateexception.py <<EOF
   > from mercurial import (
   >   error,
   >   extensions,
   >   mergestate as mergestatemod,
   > )
   > 
   > def wraprecordupdates(*args):
   >     raise error.Abort(b"simulated error while recording dirstateupdates")
   > 
   > def reposetup(ui, repo):
   >     extensions.wrapfunction(mergestatemod, 'recordupdates',
   >                             wraprecordupdates)
   > EOF
 
   $ hg rm a
   $ hg commit -m 'rm a'
   $ echo "[extensions]" >> .hg/hgrc
   $ echo "dirstateex=../dirstateexception.py" >> .hg/hgrc
   $ hg up 0
   abort: simulated error while recording dirstateupdates
   [255]
   $ hg log -r . -T '{rev}\n'
   1
   $ hg status
   ? a
 
 #if dirstate-v2
 Check that folders that are prefixes of others do not throw the packer into an
 infinite loop.
 
   $ cd ..
   $ hg init infinite-loop
   $ cd infinite-loop
   $ mkdir hgext3rd hgext
   $ touch hgext3rd/__init__.py hgext/zeroconf.py
   $ hg commit -Aqm0
 
   $ hg st -c
   C hgext/zeroconf.py
   C hgext3rd/__init__.py
 
   $ cd ..
 
 Check that the old dirstate data file is removed correctly and the new one is
 valid.
 
   $ dirstate_data_files () {
   >   find .hg -maxdepth 1 -name "dirstate.*"
   > }
 
   $ find_dirstate_uuid () {
   >   hg debugstate --docket | grep uuid | sed 's/.*uuid: \(.*\)/\1/'
   > }
 
   $ find_dirstate_data_size () {
   >   hg debugstate --docket | grep 'size of dirstate data' | sed 's/.*size of dirstate data: \(.*\)/\1/'
   > }
 
   $ dirstate_uuid_has_not_changed () {
   >   # Non-Rust always rewrites the whole dirstate
   >   if [ $# -eq 1 ] || ([ -n "$HGMODULEPOLICY" ] && [ -z "${HGMODULEPOLICY##*rust*}" ]) || [ -n "$RHG_INSTALLED_AS_HG" ]; then
   >     test $current_uid = $(find_dirstate_uuid)
   >   else
   >     echo "not testing because using Python implementation"
   >   fi
   > }
 
   $ cd ..
   $ hg init append-mostly
   $ cd append-mostly
   $ mkdir dir dir2
   $ touch -t 200001010000 dir/a dir/b dir/c dir/d dir/e dir2/f dir dir2
   $ hg commit -Aqm initial
   $ hg st
   $ dirstate_data_files | wc -l
    *1 (re)
   $ current_uid=$(find_dirstate_uuid)
 
 Nothing changes here
 
   $ hg st
   $ dirstate_data_files | wc -l
    *1 (re)
   $ dirstate_uuid_has_not_changed
   not testing because using Python implementation (no-rust no-rhg !)
 
 Trigger an append with a small change to directory mtime
 
   $ current_data_size=$(find_dirstate_data_size)
   $ touch -t 201001010000 dir2
   $ hg st
   $ dirstate_data_files | wc -l
    *1 (re)
   $ dirstate_uuid_has_not_changed
   not testing because using Python implementation (no-rust no-rhg !)
   $ new_data_size=$(find_dirstate_data_size)
   $ [ "$current_data_size" -eq "$new_data_size" ]; echo $?
   0 (no-rust no-rhg !)
   1 (rust !)
   1 (no-rust rhg !)
 
 Unused bytes counter is non-0 when appending
   $ touch file
   $ hg add file
   $ current_uid=$(find_dirstate_uuid)
 
 Trigger a rust/rhg run which updates the unused bytes value
   $ hg st
   A file
   $ dirstate_data_files | wc -l
    *1 (re)
   $ dirstate_uuid_has_not_changed
   not testing because using Python implementation (no-rust no-rhg !)
 
   $ hg debugstate --docket | grep unused
   number of unused bytes: 0 (no-rust no-rhg !)
   number of unused bytes: [1-9]\d* (re) (rhg no-rust !)
   number of unused bytes: [1-9]\d* (re) (rust no-rhg !)
   number of unused bytes: [1-9]\d* (re) (rust rhg !)
 
 Delete most of the dirstate to trigger a non-append
   $ hg rm dir/a dir/b dir/c dir/d
   $ dirstate_data_files | wc -l
    *1 (re)
   $ dirstate_uuid_has_not_changed also-if-python
   [1]
 
 Check that unused bytes counter is reset when creating a new docket
 
   $ hg debugstate --docket | grep unused
   number of unused bytes: 0
 
 #endif
 
+(non-Rust always rewrites)
+
+Test the devel option to control write behavior
+==============================================
+
+Sometimes, debugging or testing the dirstate requires making sure that we have
+done a complete rewrite of the data file and have no unreachable data around,
+sometimes it requires we ensure we don't.
+
+We test the option to force this rewrite by creating the situation where an
+append would happen and check that it doesn't happen.
+
+  $ cd ..
+  $ hg init force-base
+  $ cd force-base
+  $ mkdir -p dir/nested dir2
+  $ touch -t 200001010000 f dir/nested/a dir/b dir/c dir/d dir2/e dir/nested dir dir2
+  $ hg commit -Aqm "recreate a bunch of files to facilitate append"
+  $ hg st --config devel.dirstate.v2.data_update_mode=force-new
+  $ cd ..
+
+#if dirstate-v2
+  $ hg -R force-base debugstate --docket | grep unused
+  number of unused bytes: 0
+
+Check with the option in "auto" mode
+------------------------------------
+  $ cp -a force-base append-mostly-no-force-rewrite
+  $ cd append-mostly-no-force-rewrite
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of dir on disk which will be recorded, causing a small enough change
+to warrant only an append
+
+  $ touch -t 202212010000 dir2
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=auto
+
+UUID hasn't changed and a non-zero number of unused bytes means we've appended
+
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation (no-rust no-rhg !)
+
+#if no-rust no-rhg
+The pure python implementation never appends at the time this is written.
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0 (known-bad-output !)
+#else
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: [1-9]\d* (re)
+#endif
+  $ cd ..
+
+Check the same scenario with the option set to "force-new"
+---------------------------------------------------------
+
+  $ cp -a force-base append-mostly-force-rewrite
+  $ cd append-mostly-force-rewrite
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of dir on disk which will be recorded, causing a small enough change
+to warrant only an append, but we force the rewrite
+
+  $ touch -t 202212010000 dir2
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=force-new
+
+UUID has changed and zero unused bytes means a full-rewrite happened
+
+
+#if no-rust no-rhg
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation
+#else
+  $ dirstate_uuid_has_not_changed
+  [1]
+#endif
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0
+  $ cd ..
+
+
+Check the same scenario with the option set to "force-append"
+-------------------------------------------------------------
+
+(should behave the same as "auto" here)
+
+  $ cp -a force-base append-mostly-force-append
+  $ cd append-mostly-force-append
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of dir on disk which will be recorded, causing a small enough change
+to warrant only an append, which we are forcing here anyway.
+
+  $ touch -t 202212010000 dir2
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=force-append
+
+UUID has not changed and some unused bytes exist in the data file
+
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation (no-rust no-rhg !)
+
+#if no-rust no-rhg
+The pure python implementation never appends at the time this is written.
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0 (known-bad-output !)
+#else
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: [1-9]\d* (re)
+#endif
+  $ cd ..
+
+Check with the option in "auto" mode
+------------------------------------
+  $ cp -a force-base append-mostly-no-force-rewrite
+  $ cd append-mostly-no-force-rewrite
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of everything on disk causing a full rewrite
+
+  $ touch -t 202212010005 `hg files`
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=auto
+
+UUID has changed and zero unused bytes means we've rewritten.
+
+#if no-rust no-rhg
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation
+#else
+  $ dirstate_uuid_has_not_changed
+  [1]
+#endif
+
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0 (known-bad-output !)
+  $ cd ..
+
+Check the same scenario with the option set to "force-new"
+---------------------------------------------------------
+
+(should be the same as auto)
+
+  $ cp -a force-base append-mostly-force-rewrite
+  $ cd append-mostly-force-rewrite
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of everything on disk causing a full rewrite
+
+  $ touch -t 202212010005 `hg files`
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=force-new
+
+UUID has changed and a zero number unused bytes means we've rewritten.
+
+
+#if no-rust no-rhg
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation
+#else
+  $ dirstate_uuid_has_not_changed
+  [1]
+#endif
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0
+  $ cd ..
+
+
+Check the same scenario with the option set to "force-append"
+-------------------------------------------------------------
+
+Should append even if "auto" did not
+
+  $ cp -a force-base append-mostly-force-append
+  $ cd append-mostly-force-append
+  $ current_uid=$(find_dirstate_uuid)
+
+Change mtime of everything on disk causing a full rewrite
+
+  $ touch -t 202212010005 `hg files`
+  $ hg st \
+  > --config rhg.on-unsupported=abort \
+  > --config devel.dirstate.v2.data_update_mode=force-append
+
+UUID has not changed and some unused bytes exist in the data file
+
+  $ dirstate_uuid_has_not_changed
+  not testing because using Python implementation (no-rust no-rhg !)
+
+#if no-rust no-rhg
+The pure python implementation is never appending at the time this is written.
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: 0 (known-bad-output !)
+#else
+  $ hg debugstate --docket | grep unused
+  number of unused bytes: [1-9]\d* (re)
+#endif
+  $ cd ..
+
+
+
+Get back into a state suitable for the test of the file.
+
+  $ cd ./append-mostly
+
+#else
+  $ cd ./u
+#endif
+
 Transaction compatibility
--------------------------
+=========================
 
 The transaction preserves the dirstate.
 We should make sure all of it (docket + data) is preserved
 
 #if dirstate-v2
   $ hg commit -m 'bli'
 #endif
 
   $ hg update --quiet
   $ hg revert --all --quiet
   $ rm -f a
   $ echo foo > foo
   $ hg add foo
   $ hg commit -m foo
 
 #if dirstate-v2
   $ uid=$(find_dirstate_uuid)
   $ touch bar
   $ while [ uid = $(find_dirstate_uuid) ]; do
   >    hg add bar;
   >    hg remove bar;
   > done;
   $ rm bar
 #endif
   $ hg rollback
   repository tip rolled back to revision 1 (undo commit)
   working directory now based on revision 1
 
   $ hg status
   A foo
   $ cd ..
 
 Check dirstate ordering
 (e.g. `src/dirstate/` and `src/dirstate.rs` shouldn't cause issues)
 
   $ hg init repro
   $ cd repro
   $ mkdir src
   $ mkdir src/dirstate
   $ touch src/dirstate/file1 src/dirstate/file2 src/dirstate.rs
   $ touch file1 file2
   $ hg commit -Aqm1
   $ hg st
   $ cd ..